Re: [RFC PATCH v2 3/3] tpm: of: If available use linux,sml-log to get the log and its size

[Stefan Berger]

On 3/12/24 11:43, Jarkko Sakkinen wrote:
> On Mon Mar 11, 2024 at 10:33 PM EET, Stefan Berger wrote:
> > On 3/11/24 16:25, Jarkko Sakkinen wrote:
> > > On Mon Mar 11, 2024 at 3:20 PM EET, Stefan Berger wrote:
> > > > If linux,sml-log is available use it to get the TPM log rather than the
> > > > pointer found in linux,sml-base. This resolves an issue on PowerVM and KVM
> > > > on Power where after a kexec the memory pointed to by linux,sml-base may
> > > > have become inaccessible or corrupted. Also, linux,sml-log has replaced
> > > > linux,sml-base and linux,sml-size on these two platforms.
> > > >
> > > > Keep the handling of linux,sml-base/sml-size for powernv platforms that
> > > > provide the two properties via skiboot.
> > > >
> > > > Fixes: c5df39262dd5 ("drivers/char/tpm: Add securityfs support for event log")
> > > > Signed-off-by: Stefan Berger <stefanb@xxxxxxxxxxxxx>
> > >
> > > I'm worried about not being up to date and instead using "cached" values
> > > when verifying anything from a security chip. Does this guarantee that
> > > TPM log is corrupted and will not get updated somehow?
> >
> >
> > What do you mean 'guarantee that TPM log is corrupted'?
>
> I presume that this is for power architecture but I have no idea what

Yes it is for Power. From commit message above: "This resolves an issue 
on PowerVM and KVM on Power where after a kexec the memory pointed to by 
linux,sml-base may have become inaccessible or corrupted."

> leads log being corrupted, and is the scope all of that that arch or
> some subset of CPUs.

Every CPU will see a corrupted log.

> The commit message is not very detailed on kexec scenario. It more like

I guess what is missing in the message that the buffer was not properly 
protected during the kexec and may have been overwritten for example 
since it was mistakenly assumed to be free memory?

> assumes that reader knows all the detail beforehand. So probably this
> will start to make sense once the backing story is improved, that's
> all.
>
> > The TPM was handed over from the firmware to Linux and the firmware then
> > freed all memory associated with the log and will then not create a new
> > log or touch the TPM or do anything that would require an update to the
> > handed-over log. Linux also does not append to the firmware log. So
> > whatever we now find in linux,sml-log would be the latest firmware log
> > and the state of the 'firmware PCRs' computed from it should correspond
> > to the current state of the 'firmware PCRs'.
>
> So on what CPU this happens and is there any bigger picture for that
> design choice in the firmware?

The firmware provides a call sml-handover, which hands over the TPM log 
to the caller and at the same time frees the log. You cannot call the 
firmware a 2nd time for the log.

> If it is a firmware bug, this should emit FW_BUG log message. If not,
> then this commit message should provide the necessary context.

It's not a firmware bug. The issue is that the buffer holding the TPM 
log is not properly carried across a kexec soft reboot and may for 
example have been overwritten since it was assumed to be free memory.

> BR, Jarkko