On Mon, Jul 06, 2026 at 07:20:48PM +0100, Yeoreum Yun wrote:
[...]
However, a corner case was mentioned by Sashiko:
| It appears etm4_enable_hw() modifies drvdata->ss_status while executing | via IPI, but sshot_ctrl_store() can modify the same array concurrently | from process context since the lock is no longer held across the | smp_call_function_single() call.
"drvdata->ss_status" is a race condition between SMP call and sysfs knobs. Should we change to spinlock_irqsave/irqrestore when access drvdata->ss_status?
I know this, but This is currently ignorable since whatever perf or sysfs session, the ss_status is always clear the PENDING and STATUS bits (and we make a compromise for this right now).
And the sshot_ctrl_store()'s action is always clear above bits.
So, This is ignorable comment from Sashiko.
I understand it now, we can ignore it. Thanks for explanation.