linux-kernel - Re: [PATCH v3 2/2] genirq: fasteoi supports resend on concurrent invoke

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] [day] [month] [year] [list]

Date:   Thu, 8 Jun 2023 12:12:34 +0000
From:   "Gowans, James" <jgowans@...zon.com>
To:     "maz@...nel.org" <maz@...nel.org>
CC:     "tglx@...utronix.de" <tglx@...utronix.de>,
        "Raslan, KarimAllah" <karahmed@...zon.com>,
        "liaochang1@...wei.com" <liaochang1@...wei.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "zouyipeng@...wei.com" <zouyipeng@...wei.com>,
        "chris.zjh@...wei.com" <chris.zjh@...wei.com>
Subject: Re: [PATCH v3 2/2] genirq: fasteoi supports resend on concurrent
 invoke

On Wed, 2023-06-07 at 14:13 +0100, Marc Zyngier wrote:
> On 2023-06-07 13:21, Gowans, James wrote:
> > On Tue, 2023-06-06 at 18:05 +0100, Marc Zyngier wrote:
> > > On Mon, 05 Jun 2023 16:57:23 +0100,
> > > James Gowans <jgowans@...zon.com> wrote:
> > > > ... and enable that functionality for GIC-v3 only.
> > > 
> > > nit: drop the multi-line subject.
> > 
> > Would you prefer two commits - one to introduce the functionality and
> > one
> > to enable it for GIC-v3?
> 
> I'd prefer that. It is in general better to decouple driver stuff from
> core code.

Done, new rev with cover letter here:
https://lore.kernel.org/lkml/20230608120021.3273400-1-jgowans@amazon.com/
Also added detailed testing data to the cover - I hope that's
useful/correct.

Just one more thing: I've been thinking about it more and admit to still
being unsure of the justification for when we specifically *don't* want
the resend functionality to happen: the justification for gating it behind
a flag. The example you gave was for a wake-up source where it's not
desirable for a wakeup source to be resent. But I don't see how that case
can happen in practice: either that interrupt would never get to the
handle_fast_eoi() handler (there's probably no handler to run for it?) or
if it did it would likely use different struct irq_desc per CPU - this
race would be hit all the time if wakeup sources ran handlers and shared
irq_desc.

We have something that works now so I'm happy to go with this - I just
want to point out that I'm still struggling to see when it would actually
be wrong to apply this resend logic to a user of handle_fast_eoi().

JG