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:   Thu, 11 May 2023 16:16:07 +0200
From:   Lukas Wunner <lukas@wunner.de>
To:     Lino Sanfilippo <l.sanfilippo@kunbus.com>
Cc:     Jarkko Sakkinen <jarkko@kernel.org>, oe-lkp@lists.linux.dev,
        lkp@intel.com, linux-kernel@vger.kernel.org,
        Michael =?iso-8859-1?Q?Niew=F6hner?= <linux@mniewoehner.de>,
        linux-integrity@vger.kernel.org,
        Philipp Rosenberger <p.rosenberger@kunbus.com>,
        jsnitsel@redhat.com, peterz@infradead.org
Subject: Re: [linus:master] [tpm, tpm_tis] e644b2f498: RIP:acpi_safe_halt
Message-ID: <20230511141607.GA32208@wunner.de>
References: <202305041325.ae8b0c43-yujie.liu@intel.com>
 <d80b180a569a9f068d3a2614f062cfa3a78af5a6.camel@kernel.org>
 <42ea93a1-3186-b8ff-c317-d51f3e13856e@kunbus.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <42ea93a1-3186-b8ff-c317-d51f3e13856e@kunbus.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Thu, May 11, 2023 at 01:22:19PM +0200, Lino Sanfilippo wrote:
> Since beside the one reported by Peter Zijlstra 
> (https://lore.kernel.org/linux-integrity/CSJ0AD1CFYQP.T6T68M6ZVK49@suppilovahvero/T/#t)
> we have another interrupt storm here, it is probably the best to handle those in general
> and to disable interrupts in this case to fall back to polling (this is also what Jerry
> suggested in the thread above).
> 
> I will try to provide a patch for this.

In tpm_tis_probe_irq_single(), after you've requested the irq,
you could convert it to a struct irq_desc (via irq_to_desc()
from <linux/irqnr.h>) and cache that pointer in priv.

Then in tis_int_handler(), you could access the irqs_unhandled
member of struct irq_desc (from <linux/irqdesc.h>) and check
if it exceeds, say, 5000.

If it does, schedule a work_struct which calls disable_interrupts().
You can't call that from the IRQ handler because devm_free_irq()
waits for the IRQ handler to finish, so you'd deadlock.  You *can*
of course clear the TPM_GLOBAL_INT_ENABLE bit from the IRQ handler,
though it's unclear to me if that's sufficient to quiesce the
interrupt line.

By reusing the genirq subsystem's irqs_unhandled infrastructure,
you avoid having to reimplement all of that.

Thanks,

Lukas