Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <59d5026897bff15e371f2770335270cf1b540766.camel@kernel.org>
Subject: Re: [PATCH v3 0/4] Fixes for TPM interrupt handling
From:   Jarkko Sakkinen <jarkko@kernel.org>
To:     Lino Sanfilippo <LinoSanfilippo@gmx.de>,
        Michael =?ISO-8859-1?Q?Niew=F6hner?= <linux@mniewoehner.de>
Cc:     peterhuewe@gmx.de, jgg@ziepe.ca, stefanb@linux.vnet.ibm.com,
        stefanb@linux.ibm.com, James.Bottomley@hansenpartnership.com,
        keescook@chromium.org, jsnitsel@redhat.com, ml.linux@elloe.vision,
        linux-integrity@vger.kernel.org, linux-kernel@vger.kernel.org,
        twawrzynczak@chromium.org
Date:   Wed, 20 Apr 2022 08:32:19 +0300
In-Reply-To: <c8c0c5fb614d8b2de2a5faee2ef5ff3214281064.camel@kernel.org>
References: <20210501135727.17747-1-LinoSanfilippo@gmx.de>
         <20210501135727.17747-3-LinoSanfilippo@gmx.de>
         <YJAby8mmiJ74qWAh@kernel.org> <6722bf6f-1a3f-ee9c-55e2-cf63c64266a9@gmx.de>
         <YJNKs8bUMGOzFre+@kernel.org>
         <2a1a1cf61732eff1608aeae74054a0c135c1671f.camel@mniewoehner.de>
         <Yj0lhqTP1RoedxSc@iki.fi>
         <0d6c22b40a2f17d4b260f287d4c479a96a88b0b1.camel@mniewoehner.de>
         <efdb99b3-6d33-38b1-64a0-671821101631@gmx.de>
         <c8c0c5fb614d8b2de2a5faee2ef5ff3214281064.camel@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.44.0 
MIME-Version: 1.0
Precedence: bulk

On Wed, 2022-04-20 at 08:30 +0300, Jarkko Sakkinen wrote:
> n Sat, 2022-03-26 at 04:24 +0100, Lino Sanfilippo wrote:
> >=20
> > Hi Michael,
> >=20
> > On 25.03.22 at 13:32, Michael Niew=C3=B6hner wrote:
> > > > >=20
> > > > > Lino, I'd be happy to test the patches, when you have time and in=
terest to
> > > > > work
> > > > > on this again!
> > > > >=20
> > > > > Thanks, Michael
> > > >=20
> > > > It's quite easy to test them out. Both fixes are in the mainline GI=
T tree.
> > > > E.g. give a shot rc1, and please report if any issues persists to:
> > > >=20
> > > > =C2=A0 linux-integrity@vger.kernel.org=C2=A0
> > > >=20
> > > > BR, Jarkko
> > >=20
> > > I don't see Linos patches on mainline. Also, the series included four=
 patches:
> > > [PATCH v3 0/4] Fixes for TPM interrupt handling
> > > [PATCH v3 1/4] tpm: Use a threaded interrupt handler
> > > [PATCH v3 2/4] tpm: Simplify locality handling
> > > [PATCH v3 3/4] tpm: Fix test for interrupts
> > > [PATCH v3 4/4] tpm: Only enable supported irqs
> > >=20
> > > Three of them are relevant for the interrupt problem, which is still =
present in
> > > mainline, as these patches were refused:
> > > [PATCH v3 1/4] tpm: Use a threaded interrupt handler
> > > [PATCH v3 2/4] tpm: Simplify locality handling
> > > [PATCH v3 3/4] tpm: Fix test for interrupts
> > >=20
> > > Michael
> > >=20
> >=20
> > You are right, the interrupts are still not working in the mainline ker=
nel.
> > I would gladly make another attempt to fix this but rather step by step
> > than in a series that tries to fix (different) things at once.
> >=20
> > A first step could be to have a sleepable context for the interrupt han=
dling,
> > since in case of SPI the accesses to the irq status register may sleep.
> >=20
> > I sent a patch for this purpose once, but it seems to have gone lost:
> >=20
> > https://lore.kernel.org/all/20210620023444.14684-1-LinoSanfilippo@gmx.d=
e/
> >=20
> >=20
> > Best regards,
> > Lino
>=20
> I went these through one by one.
>=20
> # Above linked patch=20
>=20
> Boolean parameters are considered bad. I.e. use named flags
> instead. This is for above linked patch.
>=20
> # [PATCH v3 3/4] tpm: Fix test for interrupts
>=20
> 1. Please remove "unnecessarily complicated" sentence because
> =C2=A0=C2=A0 it cannot be evaluated. It's your opinion, which might perha=
ps
> =C2=A0=C2=A0 be correct, but it is irrelevant for any possible patch
> =C2=A0=C2=A0 description.
> 2. There's no such thing as "fix by re-implementation". Please
> =C2=A0=C2=A0 explain instead code change is relevant for the bug fix.
> 3. If set_bit() et al necessarily to fix a possible race condition
> =C2=A0=C2=A0 you need to have a separate patch for that.
>=20
> To move forward, start with a better summary such as
>=20
> "tpm: move interrupt test to tpm_tis_probe_irq_single()"
>=20
> I'd also either revert the change for flags, or alternatively
> move it to separate patch explaining race condition. Otherwise,
> there's no argument of saying that using set_bit() is more=20
> proper. This will make the change more localized.
>=20
>=20
> # [PATCH v3 2/4] tpm: Simplify locality handling
>=20
> "As a side-effect these modifications fix a bug which results in the
> following warning when using TPM 2:"
>=20
> Generally speaking, the simplifications should be done on top of code
> that does not have known bugs, even if the simplification renders out
> the bug. This is because then we have code that have potentially unknown
> unknown bugs.
>=20
> I hope you see my point. The problem with these patches were then
> and is still that they intermix bug fixes and other modifications and
> thus cannot be taken in.

I.e. to move forward create first localized fixes, and only after those
clean ups if there is point. Removing code (like in 2/4) is not a bug
fix fo anything. This not to say that some changes would be illegit, I'm
only saying that the patches are badly scoped.

BR, Jarkko