Re: linuxnext-2019119 edac warns (was Re: edac KASAN warning in experimental arm64 allmodconfig boot)

[John Garry]

On 22/11/2019 11:28, Robert Richter wrote:
> On 21.11.19 15:23:42, John Garry wrote:
> > On 21/11/2019 14:23, Robert Richter wrote:
> > > On 21.11.19 12:34:22, John Garry wrote:
>
> > > > [   22.046666] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.046666]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.058311] ghes_edac: Can't register at EDAC core
> > > > [   22.065402] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.065402]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.077080] ghes_edac: Can't register at EDAC core
> > > > [   22.084140] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.084140]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.095789] ghes_edac: Can't register at EDAC core
> > > > [   22.102873] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.102873]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.115442] ghes_edac: Can't register at EDAC core
> > > > [   22.122536] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.122536]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.134344] ghes_edac: Can't register at EDAC core
> > > > [   22.141441] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.141441]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.153089] ghes_edac: Can't register at EDAC core
> > > > [   22.160161] EDAC MC: bug in low-level driver: attempt to assign
> > > > [   22.160161]     duplicate mc_idx 0 in add_mc_to_global_list()
> > > > [   22.171810] ghes_edac: Can't register at EDAC core
> > >
> > > What I am more concerned is this here. In total this implies 8 ghes
> > > users that all try to register a (single-instance) ghes mc device. For
> > > non-x86 only one instance is allowed (see ghes_edac_register(), idx =
> > > 0).
>
> I also looked into this: With refcount_inc_checked() enabled, the
> refcount is *not* increased from 0 to 1. 

Yeah, I had quickly checked this back then and I think you're right.

Thanks,
John

Under the hood only
> refcount_inc_not_zero() is called instead of refcount_inc(). So the
> refcount is still zero after an edac mc device was registered. Instead
> of sharing the edac mc device, the driver tries to allocate another mc
> device for each GHESv2 entry in the HEST table. This causes the
> 'duplicate mc_idx' message. Also, it is ok to have multiple GHESv2
> entries (your system seems to have 8 entries), e.g. to serve different
> kind of errors in the system.
>
> Thanks,
>
> -Robert
> .