Re: [PATCH] tpm: ibmvtpm: Call tpm2_sessions_init() to initialize session support

["Jarkko Sakkinen" <jarkko.sakkinen@iki.fi>]

On Mon Jul 1, 2024 at 6:29 PM UTC, Stefan Berger wrote:
>
>
> On 7/1/24 11:22, Jarkko Sakkinen wrote:
> > On Fri, 2024-06-28 at 17:00 +0200, Linux regression tracking (Thorsten Leemhuis) wrote:
> >> [CCing the regression list]
> >>
> >> On 20.06.24 00:34, Stefan Berger wrote:
> >>> Jarkko,
> >>>    are you ok with this patch?
> >>
> >> Hmmm, hope I did not miss anythng, but looks like nothing happened for
> >> about 10 days here. Hence:
> >>
> >> Jarkko, looks like some feedback from your side really would help to
> >> find a path to get this regression resolved before 6.10 is released.
> >>
> >> Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat)
> > 
> > Sorry for latency, and except a bit more slow phase also during
> > July because I'm most of this month on Holiday, except taking care
> > 6.11 release.
> > 
> > This really is a bug in the HMAC code not in the IBM driver as
> > it should not break because of a new feature, i.e. this is only
> > correct conclusions, give the "no regressions" rule.
> > 
> > Since HMAC is by default only for x86_64 and it does not break
> > defconfig's, we should take time and fix the actual issue.
>
> It was enabled it on my ppc64 system after a git pull -- at least I did 
> not enable it explicitly. Besides that others can enable it on any arch 
> unless you now change the 'default x86_64' to a 'depends x86_64' iiuc 
> otherwise the usage of a Fixes: , as I used in my patch, would be justified.
>
> config TCG_TPM2_HMAC
> 	bool "Use HMAC and encrypted transactions on the TPM bus"
> 	default X86_64
> 	select CRYPTO_ECDH
> 	select CRYPTO_LIB_AESCFB
> 	select CRYPTO_LIB_SHA256
>
> https://elixir.bootlin.com/linux/v6.10-rc6/source/drivers/char/tpm/Kconfig

Yep, it is still a bug, and unmodified IBM vtpm driver must be expected
to work. I was merely saying that there is some window to  fix it properly
instead of duct tape since it is not yet widely enable feature.

I was shocked to see that the implementation has absolutely no checks
whether chip->auth was allocated. I mean anything that would cause
tpm2_sessions_init() not called could trigger null dereference.

So can you test this and see how your test hardware behaves:

https://lore.kernel.org/linux-integrity/20240701170735.109583-1-jarkko@kernel.org/T/#u

I'll modify it accrodingly if problems persist. Please put your feedback
over there. I cannot anything but compile test so it could be that
I've ignored something.

BR, Jarkko