Re: [PATCH 02/18] block: open code kobj_map into in block/genhd.c

[Christoph Hellwig]

On Fri, Oct 30, 2020 at 11:40:33AM +0100, Greg Kroah-Hartman wrote:
> On Thu, Oct 29, 2020 at 08:32:42PM +0100, Christoph Hellwig wrote:
> > On Thu, Oct 29, 2020 at 08:22:36PM +0100, Greg Kroah-Hartman wrote:
> > > After this, you want me to get rid of kobj_map, right?  Or you don't
> > > care as block doesn't use it anymore?  :)
> > 
> > I have a patch to kill it, but it causes odd regressions with the
> > tpm driver according to the kernel test.  As I have grand plans that
> > build on the block ѕide of this series for 5.11, I plan to defer the
> > chardev side and address it for 5.12.
> 
> Ok, sounds good.
> 
> Wow, I just looked at the tpm code, and it is, um, "interesting" in how
> it thinks device lifespans work.  Nothing like having 4 different
> structures with different lifespans embedded within a single structure.
> Good thing that no one can dynamically remove a TPM device during
> "normal" operation.

The regressions were during suspend then the tpm gets removed.  In
fact I'm pretty sure it is an existing problem that the change in the
lookup just surfaced in a way that the test bot notices, but I didn't
want to guard the block changes on it.