Re: [RFC PATCH] resource: Fix CXL node not populated issue

[Andy Shevchenko]

On Wed, Dec 04, 2024 at 10:07:16AM +0800, Huang, Ying wrote:
> Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> writes:
> > On Tue, Dec 03, 2024 at 02:26:52PM +0800, Huang, Ying wrote:
> >> Raghavendra K T <raghavendra.kt@xxxxxxx> writes:

...

> >> > git bisect had led to below commit
> >> > Fixes: b4afe4183ec7 ("resource: fix region_intersects() vs add_memory_driver_managed()")
> >> 
> >> This breaks you case, sorry about that.  But this also fixed a real bug
> >> too.  So, it's not appropriate just to revert it blindly.
> >
> > Linus was clear about this recently. Even if it fixes a bug, regression is
> > still regression and might (*) lead to a revert.
> > https://lwn.net/Articles/990599/
> >
> > (*) in general fixes are better than reverts, but depends on the timing in
> >     the release cycle the revert may be the only option.
> 
> I don't think that the timing is so tight that we should not work on
> proper fix firstly.  I'm trying to work with the reporter on this.

I agree on this, please do.

> BTW, the commit b4afe4183ec7 ("resource: fix region_intersects() vs
> add_memory_driver_managed()") fixed a security related bug.  The bug
> weakened the protection to prevent users read/write system memory via
> /dev/mem.  So, IMO, we need to be more careful about this.

My point was that the regression is obvious and it needs to be fixed.
That's all. Revert is a last resort in this sense.

-- 
With Best Regards,
Andy Shevchenko