[6.4-rc6] Crash during a kexec operation (tpm_amd_is_rng_defective)

[Limonciello, Mario]

[Public]

> -----Original Message-----
> From: Jerry Snitselaar <jsnitsel at redhat.com>
> Sent: Thursday, June 29, 2023 12:07 PM
> To: Limonciello, Mario <Mario.Limonciello at amd.com>
> Cc: Michael Ellerman <mpe at ellerman.id.au>; Linux regressions mailing list
> <regressions at lists.linux.dev>; Sachin Sant <sachinp at linux.ibm.com>; open
> list <linux-kernel at vger.kernel.org>; linuxppc-dev <linuxppc-
> dev at lists.ozlabs.org>; jarkko at kernel.org; linux-integrity at vger.kernel.org
> Subject: Re: [6.4-rc6] Crash during a kexec operation
> (tpm_amd_is_rng_defective)
>
> On Thu, Jun 22, 2023 at 09:38:04AM -0500, Limonciello, Mario wrote:
> >
> > On 6/22/2023 7:36 AM, Michael Ellerman wrote:
> > > "Linux regression tracking (Thorsten Leemhuis)"
> <regressions at leemhuis.info> writes:
> > > > Hi, Thorsten here, the Linux kernel's regression tracker. Top-posting
> > > > for once, to make this easily accessible to everyone.
> > > >
> > > > As Linus will likely release 6.4 on this or the following Sunday a quick
> > > > question: is there any hope this regression might be fixed any time
> > > > soon?
> > > No.
> > >
> > > I have added the author of the commit to Cc, maybe they can help?
> > >
> > > The immediate question is, is it expected for chip->ops to be NULL in
> > > this path? Obviously on actual AMD systems that isn't the case,
> > > otherwise the code would crash there. But is the fact that chip->ops is
> > > NULL a bug in the ibmvtpm driver, or a possibility that has been
> > > overlooked by the checking code.
> > >
> > > cheers
> >
> > All that code assumes that the TPM is still functional which
> > seems not to be the case for your TPM.
> >
> > This should fix it:
> >
> > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
> > index 5be91591cb3b..7082b031741e 100644
> > --- a/drivers/char/tpm/tpm-chip.c
> > +++ b/drivers/char/tpm/tpm-chip.c
> > @@ -525,6 +525,9 @@ static bool tpm_amd_is_rng_defective(struct
> tpm_chip
> > *chip)
> >         u64 version;
> >         int ret;
> >
> > +       if (!chip->ops)
> > +               return false;
> > +
> >         if (!(chip->flags & TPM_CHIP_FLAG_TPM2))
> >                 return false;
>
>
> Should tpm_amd_is_rng_defective compile to nothing on non-x86
> architectures? This code is all about
> working around an issue with the AMD fTPM, right?
>

That's a good point.  Yes it could and that would also solve this problem.