Re: [PATCH] cxl/region: Hold memdev lock during region poison injection/clear

[Li Ming]

在 2026/3/19 22:47, Dave Jiang 写道:
> On 3/19/26 7:12 AM, Li Ming wrote:
> > cxl_dpa_to_region() has expectations that cxlmd->endpoint remains valid
> > for the duration of the call. When userspace performs poison injection
> > or clearing on a region via debugfs, holding cxl_rwsem.region and
> > cxl_rwsem.dpa alone is insufficient, these locks do not prevent the
> > retrieved CXL memdev from being destroyed, nor do they protect against
> > concurrent driver detachment. Therefore, hold CXL memdev lock in the
> > debugfs callbacks to ensure the cxlmd->dev.driver remains stable for the
> > entire execution of the callback functions.
> >
> > To keep lock sequence(cxlmd.dev -> cxl_rwsem.region -> cxl_rwsem.dpa)
> > for avoiding deadlock. the interfaces have to find out the correct CXL
> > memdev at first, holding lock in the sequence then checking if the DPA
> > data has been changed before holding locks.
> >
> > Suggested-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > Signed-off-by: Li Ming <ming.li@xxxxxxxxxxxx>
> This is the patch I dropped right? I see the review tags are dropped. Does it need to be re-reviewed? Are there new changes?
>
> DJ

It is not, the patch you dropped is for ensuring memdev poison 
injection/clearing would not access an invalid cxlmd->endpoint.

The changes in this patch is for region poison injection/clearing.

But you can keep dropping the patch you mentioned at present, I will 
tell you after confirming the solution.


Thanks

Ming