Re: [PATCH v1] Revert "tracing: Fix tracing_marker may trigger page fault during preempt_disable"

On Tue, 7 Oct 2025 11:19:34 -0700
Runping Lai [off-list ref] wrote:

Hey Steve,

On Pixel6, CONFIG_PREEMPT is set. And CONFIG_PREEMPT_NONE
is not set. I'll paste the full PREEMPT configs:

Hmm, now I'm curious to why _nofault() does't work but inatomic does.

quoted

The original code is buggy, but if this is causing a regression, then we
likely need to do something else, like copy in a pre-allocated buffer?  

Sounds like a good plan. Before the long term fix, can we please
revert this commit?

Not that easy as the old way has a legitimate bug in it.

Could you see if adding pagefault_disable() and pagefault_enable() around
the __copy_from_user_inatomic() works for you? Or does that have the same
problem?

I do have a working solution, but it isn't trivial.

-- Steve