Re: Patch "tpm_tis: reserve chip for duration of tpm_tis_core_init" has been added to the 5.4-stable tree

[Dan Williams]

On Mon, Dec 30, 2019 at 4:30 PM Jerry Snitselaar <jsnitsel@xxxxxxxxxx> wrote:
>
> On Sun Dec 29 19, Dan Williams wrote:
> >On Sat, Dec 28, 2019 at 9:17 AM Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
> >>
> >> On Sat, Dec 28, 2019 at 7:15 AM Jarkko Sakkinen
> >> <jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
> >> >
> >> > On Fri, Dec 27, 2019 at 08:11:50AM +0200, Jarkko Sakkinen wrote:
> >> > > Dan, please also test the branch and tell if other patches are needed.
> >> > > I'm a bit blind with this as I don't have direct access to the faulting
> >> > > hardware. Thanks. [*]
> >> > >
> >> > > [*] https://lkml.org/lkml/2019/12/27/12
> >> >
> >> > Given that:
> >> >
> >> > 1. I cannot reproduce the bug locally.
> >> > 2. Neither of the patches have any appropriate tags (tested-by and
> >> >    reviewed-by). [*]
> >> >
> >> > I'm sorry but how am I expected to include these patches?
> >>
> >> Thanks for the branch, I'll get it tested on the failing hardware.
> >> Might be a few days due to holiday lag.
> >
> >This looked like the wrong revert to me, and testing confirms that
> >this does not fix the problem.
> >
> >As I mentioned in the original report [1] the commit that bisect flagged was:
> >
> >    5b359c7c4372 tpm_tis_core: Turn on the TPM before probing IRQ's
> >
> >That commit moved tpm_chip_start() before irq probing. Commit
> >21df4a8b6018 "tpm_tis: reserve chip for duration of tpm_tis_core_init"
> >does not appear to change anything in that regard.
> >
> >Perhaps this hardware has always had broken interrupts and needs to be
> >quirked off? I'm trying an experiment with tpm_tis_core.interrupts=0
> >workaround.
> >
>
> Hi Dan,
>
> Just to make sure I understand correctly are you saying you still have
> the screaming interrupt with the flag commit reverted,

Correct.

> or that it is
> polling instead of using interrupts [2]? Was that testing with both
> commits reverted, or just the flag commit?

With both patches reverted the driver falls back to polled mode, with
just the flag commit reverted the screaming interrupt issue is still
present.

> What kernel were you
> running before you saw the issue with 5.3 stable?

The regression was detected when moving to v5.3.6 which includes
commit 7f064c378e2c "tpm_tis_core: Turn on the TPM before probing
IRQ's".

> On that kernel you
> weren't seeing the polling message, and interrupts were working?

I've never seen interrupts working.

> Are
> you able to boot a 5.0 kernel on the system? It would be interesting
> to see how it was behaving before the power gating changes. I think it
> would be using polling due to how the code behaves because of that
> flag. It looks like without the flag being enabled by Stefan's commit
> TPM_GLOBAL_INT_ENABLE will never get cleared because tpm_tis_probe_irq_single
> expects tpm_tis_send to clear it if there is a problem, and without the
> flag being set that whole section of code is skipped.

I'll try to get a result from a pre-5.3.4 kernel to see what the
behavior is. I did have system owner run an experiment with
tpm_tis.interrupts=0 on the kernel command line and that also avoids
the problem.