From:   Thomas Gleixner <tglx@linutronix.de>
To:     Robin Murphy <robin.murphy@arm.com>, Will Deacon <will@kernel.org>
Cc:     Nishanth Menon <nm@ti.com>, Mark Rutland <mark.rutland@arm.com>,
        Stuart Yoder <stuyoder@gmail.com>, linux-pci@vger.kernel.org,
        Ashok Raj <ashok.raj@intel.com>, Marc Zygnier <maz@kernel.org>,
        x86@kernel.org, Sinan Kaya <okaya@kernel.org>,
        iommu@lists.linux-foundation.org,
        Bjorn Helgaas <helgaas@kernel.org>,
        Megha Dey <megha.dey@intel.com>,
        Jason Gunthorpe <jgg@nvidia.com>,
        Kevin Tian <kevin.tian@intel.com>,
        Alex Williamson <alex.williamson@redhat.com>,
        Santosh Shilimkar <ssantosh@kernel.org>,
        linux-arm-kernel@lists.infradead.org,
        Tero Kristo <kristo@kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Vinod Koul <vkoul@kernel.org>, dmaengine@vger.kernel.org
Subject: Re: [patch 33/37] iommu/arm-smmu-v3: Use msi_get_virq()
In-Reply-To: <76a1b5c1-01c8-bb30-6105-b4073dc23065@arm.com>
References: <20211126224100.303046749@linutronix.de>
 <20211126230525.885757679@linutronix.de>
 <20211129105506.GA22761@willie-the-truck>
 <76a1b5c1-01c8-bb30-6105-b4073dc23065@arm.com>
Date:   Mon, 29 Nov 2021 15:42:14 +0100
Message-ID: <87czmjdnw9.ffs@tglx>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On Mon, Nov 29 2021 at 13:13, Robin Murphy wrote:
> On 2021-11-29 10:55, Will Deacon wrote:
>>> -	}
>>> +	smmu->evtq.q.irq = msi_get_virq(dev, EVTQ_MSI_INDEX);
>>> +	smmu->gerr_irq = msi_get_virq(dev, GERROR_MSI_INDEX);
>>> +	smmu->priq.q.irq = msi_get_virq(dev, PRIQ_MSI_INDEX);
>> 
>> Prviously, if retrieval of the MSI failed then we'd fall back to wired
>> interrupts. Now, I think we'll clobber the interrupt with 0 instead. Can
>> we make the assignments to smmu->*irq here conditional on the MSI being
>> valid, please?
>
> I was just looking at that too, but reached the conclusion that it's 
> probably OK, since consumption of this value later is gated on 
> ARM_SMMU_FEAT_PRI, so the fact that it changes from 0 to an error value 
> in the absence of PRI should make no practical difference.

It's actually 0 when the vector cannot be found.

> If we don't have MSIs at all, we'd presumably still fail earlier
> either at the dev->msi_domain check or upon trying to allocate the
> vectors, so we'll still fall back to any previously-set wired values
> before getting here.  The only remaining case is if we've
> *successfully* allocated the expected number of vectors yet are then
> somehow unable to retrieve one or more of them - presumably the system
> has to be massively borked for that to happen, at which point do we
> really want to bother trying to reason about anything?

Probably not. At that point something is going to explode sooner than
later in colorful ways.

Thanks,

        tglx