[v4.0][v4.1][v4.2][Regression] tpm: fix raciness of PPI interface lookup

[v4.0][v4.1][v4.2][Regression] tpm: fix raciness of PPI interface lookup

[Joseph Salisbury]

Hi Jarkko,

A kernel bug report was opened against Ubuntu [0] and in bugzilla[1].  A
kernel bisect was performed, and it identified the following commit as
the first bad commit:

commit 0dc553652102c55a43eb1ab52e2049e478469f53
Author: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Date:   Fri Dec 12 11:46:35 2014 -0800

    tpm: fix raciness of PPI interface lookup


The regression was introduced as of v4.0-rc1.

I was hoping to get your feedback, since you are the patch author.  Do
you think gathering any additional data will help diagnose this issue,
or would it be best to submit a revert request?  An issue with the
revert request is that it would also require reverting several other
commits that came after this commit or a backport of the revert.
    

Thanks,
    
Joe

[0] http://pad.lv/1491467
[1] https://bugzilla.kernel.org/show_bug.cgi?id=53071

Re: [v4.0][v4.1][v4.2][Regression] tpm: fix raciness of PPI interface lookup

[Jason Gunthorpe]

On Mon, Sep 14, 2015 at 05:36:07PM -0400, Joseph Salisbury wrote:
> I was hoping to get your feedback, since you are the patch author.  Do
> you think gathering any additional data will help diagnose this
> issue,

Jarkko may have more to add, but from my perspective, this should have
been a no-functional change re-work patch. I'm surprised this has been
tagged, since it doesn't seem to really change much at all. I looked
again and all the code motion looks right to me.

The biggest change is that we don't call acpi_walk_namespace to get
the acpi_dev_handle, instead this now comes from pnp_acpi_device. It
would be useful to know if that is the source of the difference. ie
are the two methods are giving different answers on this hardware?

I'd also test with this line:

+       if (pnp_acpi_device(pnp_dev))
+               acpi_dev_handle = pnp_acpi_device(pnp_dev)->handle;

Changed to 'acpi_dev_handle = NULL;' - that should totally disable TPM
access to ACPI.

The ACPI code doesn't actually do anything if sysfs files are not
touched, I'm really not sure how it could be at fault.. Maybe also
comment out the sysfs_create_group, could something be touching the
files in userspace?

Honestly, I don't know enough about ACPI to guess what is going
on. Hopefully Jarkko does..

Jason

Re: [v4.0][v4.1][v4.2][Regression] tpm: fix raciness of PPI interface lookup

[Jarkko Sakkinen]

Hi

Sorry for the late reply. Looking into this now and sending feedback
later today.

/Jarkko

On Mon, Sep 14, 2015 at 05:36:07PM -0400, Joseph Salisbury wrote:
> Hi Jarkko,
> 
> A kernel bug report was opened against Ubuntu [0] and in bugzilla[1].  A
> kernel bisect was performed, and it identified the following commit as
> the first bad commit:
> 
> commit 0dc553652102c55a43eb1ab52e2049e478469f53
> Author: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> Date:   Fri Dec 12 11:46:35 2014 -0800
> 
>     tpm: fix raciness of PPI interface lookup
> 
> 
> The regression was introduced as of v4.0-rc1.
> 
> I was hoping to get your feedback, since you are the patch author.  Do
> you think gathering any additional data will help diagnose this issue,
> or would it be best to submit a revert request?  An issue with the
> revert request is that it would also require reverting several other
> commits that came after this commit or a backport of the revert.
>     
> 
> Thanks,
>     
> Joe
> 
> [0] http://pad.lv/1491467
> [1] https://bugzilla.kernel.org/show_bug.cgi?id=53071
>