The hwprobe vDSO data is populated by the first riscv_hwprobe syscall.
Some values, such as MISALIGNED_VECTOR_PERF, may depend on the async
vector unaligned access speed probe registered by
check_unaligned_access_all_cpus().

That initcall currently runs at late_initcall level. However,
rootfs_initcall enables usermode helpers before late initcalls run, so an
early helper can execute userspace and call riscv_hwprobe first.

In that case complete_hwprobe_vdso_data() consumes the initial
pending_boot_probes reference, populates the vDSO cache, and marks it
ready before the later async probe is registered. The eventual probe
result then cannot update the already-ready cache.

Move check_unaligned_access_all_cpus() to fs_initcall_sync. This still
runs after clocksource_done_booting(), so the ktime_get_mono_fast_ns()
benchmark uses a stable clocksource, but it runs before rootfs_initcall
enables usermode helpers.

Any async hwprobe probe is therefore registered before userspace can
trigger the one-time vDSO cache population.

Fixes: 6455c6c11827 ("riscv: Clean up & optimize unaligned scalar access probe")
Signed-off-by: Rui Qi <qirui.001@bytedance.com>
---
 arch/riscv/kernel/unaligned_access_speed.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/riscv/kernel/unaligned_access_speed.c b/arch/riscv/kernel/unaligned_access_speed.c
index bb57eb5d19df..ee2ac9972510 100644
--- a/arch/riscv/kernel/unaligned_access_speed.c
+++ b/arch/riscv/kernel/unaligned_access_speed.c
@@ -411,4 +411,10 @@ static int __init check_unaligned_access_all_cpus(void)
 	return 0;
 }
 
-late_initcall(check_unaligned_access_all_cpus);
+/*
+ * Run after clocksource_done_booting() so measure_cycles() uses a stable
+ * clocksource, but before rootfs_initcall() enables usermode helpers. Those
+ * helpers can reach hwprobe and populate the vDSO cache, so async hwprobe
+ * probes must be registered first.
+ */
+fs_initcall_sync(check_unaligned_access_all_cpus);
-- 
2.20.1

"Rui Qi" <qirui.001@bytedance.com> writes:
> The hwprobe vDSO data is populated by the first riscv_hwprobe syscall.
> Some values, such as MISALIGNED_VECTOR_PERF, may depend on the async
> vector unaligned access speed probe registered by
> check_unaligned_access_all_cpus().
>
> That initcall currently runs at late_initcall level. However,
> rootfs_initcall enables usermode helpers before late initcalls run, so an
> early helper can execute userspace and call riscv_hwprobe first.
>
> In that case complete_hwprobe_vdso_data() consumes the initial
> pending_boot_probes reference, populates the vDSO cache, and marks it
> ready before the later async probe is registered. The eventual probe
> result then cannot update the already-ready cache.
>
> Move check_unaligned_access_all_cpus() to fs_initcall_sync. This still
> runs after clocksource_done_booting(), so the ktime_get_mono_fast_ns()
> benchmark uses a stable clocksource, but it runs before rootfs_initcall
> enables usermode helpers.
>
> Any async hwprobe probe is therefore registered before userspace can
> trigger the one-time vDSO cache population.
>
> Fixes: 6455c6c11827 ("riscv: Clean up & optimize unaligned scalar access probe")
> Signed-off-by: Rui Qi <qirui.001@bytedance.com>

Sounds reasonable.

Reviewed-by: Nam Cao <namcao@linutronix.de>