Re: [PATCH v11 10/19] perf arch common: Use perf_env e_machine rather than arch

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Medium] is_native_compatible() incorrectly skips bitness checks for architectures that share the same e_machine (e.g., MIPS, PARISC, S390), causing 32-bit hosts to erroneously attempt native disassembly of 64-bit targets.
- [Medium] Native SPARC64 profiling fails to resolve the system objdump path due to a flawed architecture matching condition.
--

commit 0ccc4649e16f322564aa71d657e01e2dc842077f
Author: Ian Rogers <irogers@google.com>

perf arch common: Use perf_env e_machine rather than arch

This commit changes the binutils path lookup to use the e_machine ELF constant
rather than architecture string matching. It introduces is_native_compatible
to determine if the host can natively disassemble the target.

> diff --git a/tools/perf/arch/common.c b/tools/perf/arch/common.c
> --- a/tools/perf/arch/common.c
> +++ b/tools/perf/arch/common.c

[ ... ]

> @@ -141,11 +146,40 @@ static int lookup_triplets(const char *const *triplets, const char *name)
>  	return -1;
>  }
>  
> +static bool is_native_compatible(struct perf_env *env, uint16_t target, uint16_t host)
> +{
> +	if (target != host) {
> +		/* A 64-bit host can natively disassemble its 32-bit compat architecture */
> +		if (host == EM_X86_64 && target == EM_386)
> +			return true;
> +		if (host == EM_AARCH64 && target == EM_ARM)
> +			return true;
> +		if (host == EM_PPC64 && target == EM_PPC)
> +			return true;
> +		if (host == EM_SPARCV9 && target == EM_SPARC)
> +			return true;

[Severity: Medium]
Is this architecture matching condition reversed for SPARC?

When perf is built natively on a 64-bit SPARC system, EM_HOST is defined as
EM_SPARC, not EM_SPARCV9. When reading a native 64-bit SPARC perf.data file,
the target yields EM_SPARCV9.

Since host evaluates to EM_SPARC and target to EM_SPARCV9, this branch
appears unreachable.

Will this cause native SPARC64 profiling to incorrectly evaluate as a
cross-compilation environment and fail to resolve the system objdump?

> +		return false;
> +	}
> +
> +	/* target == host case */
> +	if (target == EM_RISCV) {
> +		bool target_is_64 = perf_env__kernel_is_64_bit(env);
> +		bool host_is_64 = (sizeof(void *) == 8);
> +
> +		/* 32-bit host cannot natively disassemble 64-bit target */
> +		if (!host_is_64 && target_is_64)
> +			return false;
> +	}

[Severity: Medium]
Does this bitness check need to apply to other architectures?

Architectures like MIPS, PARISC, and S390 also share the same e_machine value
across 32-bit and 64-bit binaries.

By restricting this check strictly to EM_RISCV, could a 32-bit host for these
other architectures erroneously attempt native disassembly of a 64-bit target
and fail to use the proper cross-compiler triplets?

> +
> +	return true;
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260602062452.2583619-1-irogers@google.com?part=10