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;
Date:   Mon, 16 Jan 2023 13:30:11 +0200
From:   Jarkko Sakkinen <jarkko@kernel.org>
To:     Vlastimil Babka <vbabka@suse.cz>
Cc:     Peter Huewe <peterhuewe@gmx.de>, Jason Gunthorpe <jgg@ziepe.ca>,
        Jan Dabros <jsd@semihalf.com>, regressions@lists.linux.dev,
        LKML <linux-kernel@vger.kernel.org>,
        linux-integrity@vger.kernel.org,
        Dominik Brodowski <linux@dominikbrodowski.net>,
        "Jason A. Donenfeld" <Jason@zx2c4.com>,
        Herbert Xu <herbert@gondor.apana.org.au>,
        Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [REGRESSION] suspend to ram fails in 6.2-rc1 due to tpm errors
Message-ID: <Y8U1QxA4GYvPWDky@kernel.org>
References: <7cbe96cf-e0b5-ba63-d1b4-f63d2e826efa@suse.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <7cbe96cf-e0b5-ba63-d1b4-f63d2e826efa@suse.cz>
Precedence: bulk

On Wed, Dec 28, 2022 at 09:22:56PM +0100, Vlastimil Babka wrote:
> Ugh, while the problem [1] was fixed in 6.1, it's now happening again on
> the T460 with 6.2-rc1. Except I didn't see any oops message or
> "tpm_try_transmit" error this time. The first indication of a problem is
> this during a resume from suspend to ram:
> 
> tpm tpm0: A TPM error (28) occurred continue selftest

This failure occurs in tpm_tis_resume().

tpm1_do_selftest() is called for the first time during a power cycle in
tpm_chip_register(), which does not spawn anything klog, does it?

Because they wrap the call differently tpm_tis_resume() does not disable
clkrun protocol, which is at least obvious semantical difference.

I'd suggest to workaround the bug by replacing tpm1_do_selftest() with

1. tpm_chip_start()
2. tpm1_auto_startup()
3. tpm_chip_stop()

tpm1_auto_startup() has already convergent rollback semantics as in

https://lore.kernel.org/lkml/20230105144742.3219571-1-Jason@zx2c4.com/

To add, even without bug, it makes a lot of commons to make semantics
convergent.

BR, Jarkko