On Mon, Sep 09, 2019 at 05:28:08PM +0100, Jarkko Sakkinen wrote:
> On Sat, Sep 07, 2019 at 06:04:48PM -0400, Sasha Levin wrote:
> > On Sat, Sep 07, 2019 at 09:55:18PM +0300, Jarkko Sakkinen wrote:
> > > On Tue, 2019-09-03 at 15:43 -0400, Sasha Levin wrote:
> > > > Right. I gave a go at backporting a few patches and this happens to be
> > > > one of them. It will be a while before it goes in a stable tree
> > > > (probably way after after LPC).
> > >
> > > It *semantically* depends on
> > >
> > > db4d8cb9c9f2 ("tpm: use tpm_try_get_ops() in tpm-sysfs.c.")
> > >
> > > I.e. can cause crashes without the above patch. As a code change your
> > > patch is fine but it needs the above patch backported to work in stable
> > > manner.
> > >
> > > So... either I can backport that one (because ultimately I have
> > > responsibility to do that as the maintainer) but if you want to finish
> > > this one that is what you need to backport in addition and then it
> > > should be fine.
> >
> > If you're ok with the backport of this commit, I can just add
> > db4d8cb9c9f2 on top.
>
> Sure, I've already gave my promise to do that :-)
I ended up with:
db4d8cb9c9f2 tpm: Fix TPM 1.2 Shutdown sequence to prevent future TPM operations
2677ca98ae37 tpm: use tpm_try_get_ops() in tpm-sysfs.c.
da379f3c1db0 tpm: migrate pubek_show to struct tpm_buf
Since some time has passed I'l just restate that the reason for
backporting 2677ca98ae37 was that tpm_class_shutdown() could pull carpet
under the TPM 1.2 code. tpm_try_get_ops() makes sure that read lock is
taken and chip->ops is not NULL if it successfully returns.
Still need to test the patches with TPM 1.2 hardware before I can send
them.
/Jarkko
