qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [edk2] [PATCH 6/7] ovmf: link with Tcg2ConfigDxe module


From: Stefan Berger
Subject: Re: [Qemu-devel] [edk2] [PATCH 6/7] ovmf: link with Tcg2ConfigDxe module
Date: Fri, 2 Mar 2018 08:35:21 -0500
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.4.0

On 03/02/2018 06:12 AM, Laszlo Ersek wrote:
On 03/01/18 17:59, Stefan Berger wrote:
On 02/26/2018 04:58 AM, Laszlo Ersek wrote:
On 02/23/18 14:23, address@hidden wrote:
From: Marc-André Lureau <address@hidden>

The module allows to tweak and interact with the TPM. Note that many
actions are broken due to implementation of qemu TPM (providing it's
own ACPI table), and the lack of PPI implementation.

CC: Laszlo Ersek <address@hidden>
CC: Stefan Berger <address@hidden>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Marc-André Lureau <address@hidden>
---
   OvmfPkg/OvmfPkgX64.dsc | 2 ++
   OvmfPkg/OvmfPkgX64.fdf | 1 +
   2 files changed, 3 insertions(+)

diff --git a/OvmfPkg/OvmfPkgX64.dsc b/OvmfPkg/OvmfPkgX64.dsc
index 9bd0709f98..2281bd5ff8 100644
--- a/OvmfPkg/OvmfPkgX64.dsc
+++ b/OvmfPkg/OvmfPkgX64.dsc
@@ -669,6 +669,8 @@
NULL|SecurityPkg/Library/HashInstanceLibSha1/HashInstanceLibSha1.inf NULL|SecurityPkg/Library/HashInstanceLibSha256/HashInstanceLibSha256.inf
     }
+
+  SecurityPkg/Tcg/Tcg2Config/Tcg2ConfigDxe.inf
   !endif
     !if $(SECURE_BOOT_ENABLE) == TRUE
diff --git a/OvmfPkg/OvmfPkgX64.fdf b/OvmfPkg/OvmfPkgX64.fdf
index b8dd7ecae4..985404850f 100644
--- a/OvmfPkg/OvmfPkgX64.fdf
+++ b/OvmfPkg/OvmfPkgX64.fdf
@@ -399,6 +399,7 @@ INF
MdeModulePkg/Universal/Variable/RuntimeDxe/VariableRuntimeDxe.inf
     !if $(TPM2_ENABLE) == TRUE
   INF  SecurityPkg/Tcg/Tcg2Dxe/Tcg2Dxe.inf
+INF  SecurityPkg/Tcg/Tcg2Config/Tcg2ConfigDxe.inf
   !endif
################################################################################


Please drop this patch.

In my earlier investigation I wrote, Tcg2ConfigDxe "[p]rovides a Setup
TUI interface to configure the TPM. IIUC, it can also save the
configured TPM type for subsequent boots (see Tcg2ConfigPei.inf above)".

The INF file itself says "This module is only for reference only, each
platform should have its own setup page."

And Jiewen wrote earlier, "Tcg2ConfigPei/Dxe are platform sample driver.
A platform may have its own version based upon platform requirement. For
example, if a platform supports fTPM, it may use another Tcg2Config
driver."

Given that OVMF lacks PEI-phase variable access, and that I consequently
suggested cloning, and seriously trimming, Tcg2ConfigPei, it makes no
sense to include an HII dialog that sets a variable for PEI phase
consumption. Also, as you say, many of the exposed operations are broken
due to lack of PPI support. So let's just postpone the inclusion of this
driver, for now.
Just FYI: The PPI support for the OS requires ACPI
OK

and, as it is
currently implemented, SMF where UEFI variables are manipulated.
(I assume s/SMF/SMM/)

correct.

You are correct to write "as it is currently implemented". My point in
my previous email(s) was that QEMU should generate the ACPI payload
needed by the OS, for queueing PPI opcodes (i.e. OVMF should install
QEMU's AML, *not* the sample AML code in SecurityPkg). In turn the AML
from QEMU should queue the PPI opcodes in the custom register block of
the TPM device (which is anyway the only NVRAM-emulation possibility
under SeaBIOS).

We can emulate other NVRAM as well. It is a possibility and provides the flexibility of setting flags per implemented PPI code indicating whether the firmware implements the codes and sysfs entries in Linux can then show what is actually supported. If you write code '23' into sysfs and ACPI has no clue whether the firmware implements '23', then you may reboot for nothing and start wondering what is going on because the effect isn't there. Also I don't think we'll implement the same set of commands in both firmwares that we would want to hard-code the checking for implemented code in ACPI produced by the firmware.


Given that the PPI opcodes will henceforth not be stored in a UEFI
variable under OVMF either, the SMM requirement in the AML falls away
completely.

I would not modify that code but keep what is there right now and write some glue code around it: Upon reboot detect what PPI code has been set, if any, and write it into the UEFI variable. Then existing UEFI code reads the variable (stemming either from OS or a menu operation) and acts upon it. That keeps much of the existing code unmodified, which I find appealing. Also the code for the menu wouldn't have to be modified.


Retrieving the PPI opcodes for processing from the custom MMIO register
block of the TPM device, after reboot, as opposed to reading them out of
a UEFI variable, will take custom code in OVMF. We'll get there.

That custom MMIO register block could, in the worst case, be assigned a different purpose in a future spec...

  Stefan

Some
menu items in the TPM 2 menu (also TPM 1.2) also require these UEFI
variables of the PPI interface so that UEFI can react on the menu
choices upon re.
(I assume s/re/reboot/)

In "SecurityPkg/Tcg/Tcg2Config/Tcg2Config.vfr" (which is the "Visual
Form Representation" of the dialog we're talking about), I see three
variable references. The structures for those are defined in
"SecurityPkg/Tcg/Tcg2Config/Tcg2ConfigNvData.h":

- TCG2_CONFIGURATION_INFO
- TCG2_CONFIGURATION
- TCG2_VERSION

I think the last two are irrelevant under OVMF / QEMU (fixed version 2
TPM, and ACPI comes from QEMU -- consistent with my other comments for
the PEI phase modules).

TCG2_CONFIGURATION_INFO looks more complex, so perhaps we'll have to
scavenge its handling for OVMF. However, it seems that
TCG2_CONFIGURATION_INFO is not needed in the PEI phase.

... Understand this right: I'd be bursting from joy if OVMF had
PEI-phase r/o variable access. I got that feature working for QEMU. But
all my attempts to upstream the work failed (apparently because I'm not
willing to develop a parallel "fake" for Xen -- on Xen, NVRAM/pflash
doesn't even exist, so even if the PEI variable service existed on Xen,
it would have zero chance to return valid data.)

Laszlo





reply via email to

[Prev in Thread] Current Thread [Next in Thread]