Re: [PATCH] timekeeping: publish suspend_timing_needed with release semantics

[Thomas Gleixner]

On Mon, Sep 21 2026 at 21:23, Jaidev Shastri via wrote:
> From: Jaidev Shastri <jaidevshastri@xxxxxx>
>
> timekeeping_suspend() sets suspend_timing_needed with a plain store
> after it has read the persistent clock. timekeeping_rtc_skipresume()
> reads it with a plain load on behalf of the RTC class during resume.

> Set it with smp_store_release() and read it with smp_load_acquire().
>
> Found with MBCheck, a static herd7-based memory consistency checker.

And what exactly is this release/acquire solving?

1) timekeeping_suspend()
        suspend_timing_needed = true
   ...

2) system suspends

3) system resumes

4) timekeeping_resume()
        if (inject_sleeptime)
            suspend_timing_needed = false

5) system continues to resume

6) timekeeping_rtc_skipresume()

The store in #1 is completely uninteresting because suspend/resume is a big
pile of memory barriers so there is _ZERO_ chance that #6 cannot observe
the store in #1.

The more interesting case would be the conditional store in #4 which in
a pretty far fetched theory could lead to the case that
timekeeping_rtc_skipresume() would not observe it.

But if you actually look at the sequence which is between #4 and #6 this
theory becomes a myth.

Please analyze your tool findings properly before blindly claiming that
the tool actually found something problematic.

Thanks,

        tglx