Re: [Qemu-devel] [PATCH V20 6/8] Add support for cancelling of a TPM command

[Stefan Berger]

On 02/04/2013 05:06 PM, Corey Bryant wrote:
> On 02/04/2013 01:48 PM, Stefan Berger wrote:
> > On 02/04/2013 11:02 AM, Corey Bryant wrote:
> > > > +    /*
> > > > +     * As of Linux 3.7 the tpm_tis driver does not properly cancel
> > > > +     * commands for all TPMs.
> > >
> > >
> > > Any idea what the plan is for this issue?  Is it an ongoing matter of
> > > adding kernel support as unsupported TPMs are identified?
> >
> >
> > I submitted a patch to the tpmdd-devel mailing list fixing the
> > cancellation issue with the particular TPM I have on one of my machines.
> > Kent Yoder modified it to add a fix for yet another TPM (from a
> > different vendor).
> >
> > https://github.com/shpedoikal/linux/commit/9f11986de7280f999cad342389b48c29002c0f37 
> >
> >
> >
> > The spec seems not be clear enough as to what bits must be set in the
> > status register when a TPM command is canceled so that the so far
> > implemented cancellation detection is insufficient and needs to be
> > adapted to TPM vendors' implementations. All TIS interfaces are supposed
> > to support cancellation, though.
>
> It sounds like this is a work in progress at the spec and Linux kernel 
> level.  It's certainly not ideal, but I think as long as a message is 
> issued for unsupported TPMs, it shouldn't hold up integration of this 
> support into QEMU.


There may be a misunderstanding here: whether canceling an ongoing TPM 
command works is not something that QEMU can tell you. We will always 
find the cancel sysfs entry and we can write a byte into it to trigger 
the cancellation but whether the driver then detects the cancellation 
correctly is something different - the problem here is that the thread 
is waiting inside the driver for the command to complete and 
periodically check whether it has finished and/or was canceled. I 
previously used a different machine where cancellation was working but 
now with the new machine and a different vendor's TPM the cancel sysfs 
entry cancels the command fine but the driver does not detect the fact 
that the command was actually canceled and just continues waiting for 
the command's completion. Also since we don't have access to all TPM 
vendors' TPMs we cannot tell for sure which ones are working and which 
ones are not.

   Stefan