Re: [PATCH 5/6] s390/Kconfig: Select ARCH_SUPPORTS_CFI

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Missing CFI type hashes in s390 BPF JIT causes kernel panics on valid BPF callbacks.
- [Medium] Enabling ARCH_SUPPORTS_CFI on s390 breaks the semantic contract of CFI_PERMISSIVE, causing kernel panics instead of warnings.
--

--- Patch [5]: [PATCH 5/6] s390/Kconfig: Select ARCH_SUPPORTS_CFI ---
commit 331ca22524bf97cd2604c69fc257401120f48c71
Author: Heiko Carstens <hca@linux.ibm.com>

s390/Kconfig: Select ARCH_SUPPORTS_CFI

This commit selects ARCH_SUPPORTS_CFI for s390 to enable KCFI support. It
notes that without a .kcfi_traps section, CFI mismatches will result in
kernel crashes instead of warnings, even in permissive mode.

> In addition this also means that CFI_PERMISSIVE does not work. Even if the
> option is selected the kernel will crash in case of checksum mismatch.

[Severity: Medium]
Since enabling this feature on s390 breaks the expected semantic contract
of CFI_PERMISSIVE defined in arch/Kconfig, should we actively prevent users
from selecting permissive mode on s390?

Currently, a user configuring CONFIG_CFI and CONFIG_CFI_PERMISSIVE would
expect a warning on a mismatch, but will experience a hard panic instead.

Would it be better to add a dependency restricting CFI_PERMISSIVE so it
cannot be selected when the architecture lacks trap interception?

> diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig
> index 84404e6778d50..d57a6c02f51d2 100644
> --- a/arch/s390/Kconfig
> +++ b/arch/s390/Kconfig
> @@ -144,6 +144,7 @@ config S390
>  	select ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
>  	select ARCH_STACKWALK
>  	select ARCH_SUPPORTS_ATOMIC_RMW
> +	select ARCH_SUPPORTS_CFI

[Severity: High]
Does the s390 BPF JIT implementation currently generate the required CFI type
hashes?

Looking at bpf_int_jit_compile() in arch/s390/net/bpf_jit_comp.c, if the
JIT does not yet emit the cfi_bpf_hash preceding the generated code, will
valid BPF subprogram callbacks (like bpf_task_work_callback or timers)
trigger a KCFI mismatch and crash the host?

>  	select ARCH_SUPPORTS_DEBUG_PAGEALLOC
>  	select ARCH_SUPPORTS_HUGETLBFS
>  	select ARCH_SUPPORTS_INT128 if CC_HAS_INT128 && CC_IS_CLANG

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260724141318.1037434-1-hca@linux.ibm.com?part=5