Re: [BUG] btrfs: MAX_LOCKDEP_CHAIN_HLOCKS too low!

[Mirsad Goran Todorovac]

On 6/27/23 18:27, Mirsad Goran Todorovac wrote:
> Hi, Qu, David,
>
> On 6/27/23 16:27, David Sterba wrote:
> > On Tue, Jun 27, 2023 at 08:19:37AM +0800, Qu Wenruo wrote:
> > > On 2023/6/27 07:40, Mirsad Goran Todorovac wrote:
> > > > Hi,
> > > >
> > > > There is a bug apparently in btrfs file system.
> > > >
> > > > The platform is an assembled box with Ryzen 9 processor and ASRock X670E PG
> > > > Lightning motherboard.
> > > >
> > > > I do not have a reproducer, just kernel log:
> > > >
> > > > Jun 26 20:41:58 defiant kernel: [ 2273.786736] BUG:
> > > > MAX_LOCKDEP_CHAIN_HLOCKS too low!
> > >
> > > This is really what it said.
> > >
> > > The setting is too low for certain workload.
> > >
> > > In fact Fedora is already increasing this value.
> > > If you want lockdep, I guess that's the only way to go.
> >
> > https://btrfs.readthedocs.io/en/latest/dev/Development-notes.html#bug-max-lockdep-chain-hlocks-too-low
> >
> > As the reports will probably keep coming until lockdep changes I'll move
> > the answer to a more user-visible page.
>
> Thank you for your prompt response. It wasn't obvious from the kernel
> message though which .config setting needs adjustment.
>
> I am yet unable to tell whether the lockdep setting being too low and
> turning off the locking correctness validator caused the kernel crash, but
> I will know more after restarting the kselftest with the new lockdep
> setting ...

P.S.

I heard it is uncool to answer one's own email, but I have been unable to
reproduce the crash of yesterday without or with pcscd smartcard daemon
in debug mode.

This of course doesn't mean that the LOCKDEP setting was a direct cause
of crash, I believe that findings are inconclusive. (IMHO, correct locking
should work even with locking correctness validator turned off.)

FYI, here are the crash logs, in case it could help improve btrfs:

[1] https://domac.alu.unizg.hr/~mtodorov/linux/crashes/2023-06-27/

Best regards,
Mirsad Todorovac