RE: [PATCH v4 1/2] mm/memory-failure: Use a mutex to avoid memory_failure() races

[Luck, Tony]

> Sorry to interrupt, I just thought one thing:
>
> This mutex seems not been bind to the error page, will there be some core case
> like test code or multi-poison case whick will break this mutex?

The mutex is a bigger hammer than strictly needed . If there are simultaneous
errors on different pages they could, int theory, proceed in parallel. But this mutex
will serialize the processing.

Is this a problem? Hopefully systems aren't seeing so many uncorrectable/recoverable
errors that this would be a significant bottleneck.

-Tony