linux-kernel - Re: [PATCH v3 02/13] genirq: Define ack_irq() and eoi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <877dgq9450.mognet@arm.com>
Date:   Thu, 12 Aug 2021 14:36:11 +0100
From:   Valentin Schneider <valentin.schneider@....com>
To:     Marc Zyngier <maz@...nel.org>
Cc:     linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        Thomas Gleixner <tglx@...utronix.de>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Vincenzo Frascino <vincenzo.frascino@....com>
Subject: Re: [PATCH v3 02/13] genirq: Define ack_irq() and eoi_irq() helpers

On 12/08/21 08:49, Marc Zyngier wrote:
> On Tue, 29 Jun 2021 13:49:59 +0100,
> Valentin Schneider <valentin.schneider@....com> wrote:
>> +void eoi_irq(struct irq_desc *desc)
>> +{
>> +	desc->irq_data.chip->irq_eoi(&desc->irq_data);
>> +
>> +	if (desc->irq_data.chip->flags & IRQCHIP_AUTOMASKS_FLOW)
>> +		irq_state_clr_flow_masked(desc);
>
> I just realised that this has a good chance to result in a mess with
> KVM, and specially the way we let the vGIC deactivate an interrupt
> directly from the guest, without any SW intervention (the magic HW bit
> in the LRs).
>

I didn't think to consider those. It can't ever be simple, can it...

> With this, interrupts routed to a guest (such as the timers) will
> always have the IRQD_IRQ_FLOW_MASKED flag set, which will never be
> cleared.
>
> I wonder whether this have a chance to interact badly with
> mask/unmask, or with the rest of the flow...
>

Isn't it the other way around? That is, eoi_irq() will clear
IRQD_IRQ_FLOW_MASKED regardless of what happens within chip->irq_eoi(),
so we would end up with !IRQD_IRQ_FLOW_MASKED even if the (physical) IRQ
remains Active (irqd_is_forwarded_to_vcpu() case).

This does not entirely match reality (if the IRQ is still Active then it is
still "flow-masked"), but AFAICT this doesn't impact our handling of
forwarded IRQs: IRQD_IRQ_FLOW_MASKED is only really relevant from ack_irq()
to eoi_irq(), and deactivation-from-the-guest (propagated via LR.HW=1)
happens after that.