Re: [PATCH v2] tpm: Cleanup class for tpm_buf

[Jarkko Sakkinen <jarkko@kernel.org>]

On Thu, Jun 26, 2025 at 09:11:05PM +0300, Jarkko Sakkinen wrote:
> On Thu, Jun 26, 2025 at 10:50:22AM -0400, James Bottomley wrote:
> > On Thu, 2025-06-26 at 13:19 +0300, Jarkko Sakkinen wrote:
> > > From: Jarkko Sakkinen <jarkko.sakkinen@opinsys.com>
> > > 
> > > Create a cleanup class for struct tpm_buf using DEFINE_CLASS(), which
> > > will guarantee that the heap allocated memory will be freed
> > > automatically for the transient instances of this structure, when
> > > they go out of scope.
> > > 
> > > Wrap this all into help macro CLASS_TPM_BUF().
> > > 
> > > A TPM buffer can now be declared trivially:
> > > 
> > >     CLASS_TPM_BUF(buf, buf_size);
> > 
> > Well, that's not all ... you're also adding a size to the API that we
> > didn't have before, which should at least be documented in the commit
> > message and probably be a separate patch.
> > 
> > What is the reason for this, though?  The reason we currently use a
> > page is that it's easy for the OS to manage (no slab fragmentation
> > issues).  The TCG reference platform defines this to be just under 4k
> > (actually 4096-0x80) precisely because TPM implementations don't do
> > scatter gather, so they don't want it going over an ARM page, so
> > there's no danger of us ever needing more than a page.
> 
> Thanks for the valuable feedback.
> 
> I can drop "buf_size" parameter. It is not a priority, and I also
> agree with your comments.

I also noticed that I had changed one log message in tpm2-sessions.c. It
was unintended i.e. a spurious change. I'll revert that one too.

I'll split this into more reasonable portions for next version so these
should be easier to review then.

BR, Jarkko