[edk2-devel] Problem: TPM 2.0 event log by OVMF is shown empty in Linux kernel versions after 5.8

[Laszlo Ersek]

On 04/27/21 01:44, James Bottomley wrote:
> On Mon, 2021-04-26 at 21:56 +0200, Thore Sommer wrote:
>> Dear Maintainers,
>>
>> during my testing with OVMF and swtpm I found out that kernel
>> versions newer than 5.8 don't show any information in 
>> "/sys/kernel/security/tpm0/binary_bios_measurements" if swtpm
>> emulates a TPM 2.0 device. The file is still created but is empty.
>> The expected result would be that 
>> "/sys/kernel/security/tpm0/binary_bios_measurements" contains the
>> TPM event log. TPM 1.2 devices are not affected.
> 
> I don't confirm this.  I have Linux version 5.12.0-rc5+ installed and I
> see the attached in my binary_bios_measurements (I've run it through
> tpm2-eventlog so you can see the actual events).
> 
>> With the help of git bisect I found out that the breaking kernel
>> commit is 85467f63a05c43364ba0b90d0c05bb89191543fa.
>> Reverting this on top the 5.12 release restores the expected
>> functionality.
>>
>> Thanks to apalos and leiflindholm on the #edk2 IRC channel for
>> helping me with that.
>>
>> I don't know if this is a bug in OVMF or in the Linux kernel, because
>> on a real device with a TPM 2.0 the output was as expected.
>>
>> Tested with edk2-ovmf 202102, swtpm 0.5.2 and qemu 5.2.0 on Ubuntu
>> 20.04.
>>
>> If further information is needed to resolve this problem, I'd be
>> happy to provide them.
> 
> What that commit did was to allow the event log to be provided by the
> ACPI table if one existed rather than always defaulting to it being
> provided by the EFI configuration table.  What I suspect has happened
> from this:
> 
>> [    0.017358] ACPI: Reserving TPM2 table memory at [mem 
>> 0x7eb77000-0x7eb7704b]
> 
> Is that somehow you've got an empty TPM2 table installed in ACPI but I
> don't know how you've done this.  On my OVMF boot I'm using the direct
> kernel command line and I have secure boot enabled but not activated,
> which is why you only see PCRs 0-7 in the log.

IIRC the QEMU ACPI linker/loader exposes a TPM2 ACPI table as well;
maybe that conflicts with the edk2 TPM2 machinery built into OVMF,
somehow. An OVMF log (enabling DEBUG_VERBOSE) might help. Running
acpidump + iasl in the guest might help as well (for determining some
inconsistency). My gut feeling is that it's a fight between QEMU's ACPI
generator and the edk2 TPM infrastructure, over the ownership of the
TPM-related ACPI table(s).

Laszlo



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#74531): https://edk2.groups.io/g/devel/message/74531
Mute This Topic: https://groups.io/mt/82391340/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-