On 6/8/26 10:50, Steven Rostedt wrote:
On Sun, 7 Jun 2026 15:24:31 +0800
Hui Wang <[email protected]> wrote:
trace_marker_raw.tc assumes that the raw marker payload length
reported in trace_pipe is the result of int((id + 3) / 4) * 4, but
that is not true on kernels with CONFIG_HAVE_64BIT_ALIGNED_ACCESS
enabled.
With forced 8-byte alignment, the ring buffer event forces 8-byte
alignment. The event length is stored in array[0], the payload data
and id are placed in a struct raw_data_entry which is stored starting
at array[1]. In this case, the printed payload data length is 8*N+4
bytes.
To make the testcase pass in this case, add a kconfig_enabled() helper
and use it to detect CONFIG_HAVE_64BIT_ALIGNED_ACCESS so
trace_marker_raw.tc can calculate the expected length correctly.
Assisted-by: Copilot:gpt-5.5
Signed-off-by: Hui Wang <[email protected]>
NACK
Let's not change the kernel for a broken test. Also this has already
been fixed but appears not to be applied yet.
Shuah, can you please apply the below fix.
https://lore.kernel.org/all/[email protected]/
I applied the above to linux-kselftest next - will send it up later
this week to for Linux 7.2-rc1
thanks,
-- Shuah
