Re: [PATCH] firmware: qcom: scm: Hide QCOM_SCM instead of depending on ARCH_QCOM

From: Nathan Chancellor

Date: Sun Nov 22 2026 - 00:16:46 EST


Hi Sasha and qcom folks,

On Sun, Sep 13, 2026 at 01:30:49PM -0400, Sasha Levin wrote:
> imx_v6_v7_defconfig fails to link:
>
> arm-linux-gnueabihf-ld: drivers/firmware/qcom/qcom_scm.o: in function
> `qcom_scm_pas_init_image':
> qcom_scm.c:(.text+0xecc): undefined reference to `qcom_tzmem_alloc'
> qcom_scm.c:(.text+0x2b0): undefined reference to `qcom_tzmem_free'
> qcom_scm.c:(.text+0x1514): undefined reference to `qcom_tzmem_to_phys'
>
> QCOM_SCM gained a "depends on ARCH_QCOM || COMPILE_TEST", but it is
> selected from 20 places in the tree and several of those are reachable
> without ARCH_QCOM. DRM_MSM is one of them, via SOC_IMX5, which is how
> imx_v6_v7_defconfig ends up here:
>
> WARNING: unmet direct dependencies detected for QCOM_SCM
> Depends on [n]: ARCH_QCOM [=n] || COMPILE_TEST [=n]
> Selected by [y]:
> - DRM_MSM [=y] && ... && (ARCH_QCOM [=n] || SOC_IMX5 [=y] ||
> COMPILE_TEST [=n]) && ...
>
> kconfig still honours the select and sets QCOM_SCM=y, but it does not
> propagate the selects of a symbol whose own dependencies are unmet, so
> QCOM_TZMEM is left unset, qcom_tzmem.c is not built, and the
> unconditional qcom_tzmem_*() calls in qcom_scm.c and qcom_scm-smc.c
> have nothing to bind to.
>
> The point of the dependency was only to stop asking the user about
> Qualcomm firmware interfaces when configuring a non-Qualcomm kernel.
> Express that as a prompt condition, which is what the same commit
> already did for QCOM_PAS: the symbol stays invisible, but it no longer
> carries a dependency that silently drops its selects.
>
> Found by KernelCI builds of the linus-next tree.
>
> Fixes: 8fecd3194de3 ("firmware: QCOM interfaces should depend on ARCH_QCOM")
> Reported-by: kernel test robot <lkp@xxxxxxxxx>
> Closes: https://lore.kernel.org/oe-kbuild-all/202609120634.g0HMAp3g-lkp@xxxxxxxxx/
> Assisted-by: LLM
> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
> ---
> drivers/firmware/qcom/Kconfig | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/firmware/qcom/Kconfig b/drivers/firmware/qcom/Kconfig
> index 691db0ded8e8c..b50e1309ced5f 100644
> --- a/drivers/firmware/qcom/Kconfig
> +++ b/drivers/firmware/qcom/Kconfig
> @@ -26,8 +26,7 @@ config QCOM_PAS_TEE
> by the firmware TEE implementation as the backend.
>
> config QCOM_SCM
> - tristate "Qualcomm PAS SCM interface driver"
> - depends on ARCH_QCOM || COMPILE_TEST
> + tristate "Qualcomm PAS SCM interface driver" if ARCH_QCOM || COMPILE_TEST
> select QCOM_PAS
> select QCOM_TZMEM
> default y if ARCH_QCOM

This now allows

# CONFIG_ARCH_QCOM is not set
CONFIG_QCOM_SCM=m
CONFIG_QCOM_TZMEM=m
# CONFIG_QCOM_TZMEM_MODE_GENERIC is not set
CONFIG_QCOM_TZMEM_MODE_SHMBRIDGE=y

which breaks the build (or install if depmod is postponed until then)
for me. I ran into this slimming down platform support from my aarch64
Fedora kernel, as I don't have any Qualcomm platforms.

$ cat allno.config
CONFIG_ATH12K=m
CONFIG_ATH12K_AHB=y
CONFIG_CFG80211=m
CONFIG_MAC80211=m
CONFIG_MODULES=y
CONFIG_NET=y
CONFIG_NETDEVICES=y
CONFIG_PCI=y
CONFIG_QCOM_TZMEM_MODE_GENERIC=n
CONFIG_QCOM_TZMEM_MODE_SHMBRIDGE=y
CONFIG_REMOTEPROC=y
CONFIG_WLAN=y
CONFIG_WLAN_VENDOR_ATH=y

$ make -skj"$(nproc)" \
ARCH=arm64 \
CROSS_COMPILE=aarch64-linux-gnu- \
KCONFIG_ALLCONFIG=1 \
INSTALL_MOD_PATH=$PWD/rootfs \
clean allnoconfig all modules_install
depmod: ERROR: Cycle detected: qcom_scm -> qcom_tzmem -> qcom_scm
depmod: ERROR: Found 2 modules in dependency cycles!

One way around this is turning CONFIG_QCOM_SCM into a bool but that
would end up building more code into the kernel image that may or may
not run. The proper solution is probably doing what Geert suggested and
only selecting these symbols with ARCH_QCOM but I did not explore this.

--
Cheers,
Nathan