linux-kernel - Re: [PATCH v4] ACPI: AGDI: Add interrupt signaling mode support

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 for Android: free password hash cracker in your pocket

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

Message-ID: <aRRYW7UK7PWb6cpS@lpieralisi>
Date: Wed, 12 Nov 2025 10:50:19 +0100
From: Lorenzo Pieralisi <lpieralisi@...nel.org>
To: Charles Mirabile <cmirabil@...hat.com>
Cc: fj1078ii@...itsu.com, catalin.marinas@....com, fj2767dz@...itsu.com,
	guohanjun@...wei.com, ilkka@...amperecomputing.com, lenb@...nel.org,
	linux-acpi@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
	linux-kernel@...r.kernel.org, rafael@...nel.org,
	sudeep.holla@....com, will@...nel.org, tglx@...utronix.de,
	maz@...nel.org
Subject: Re: [PATCH v4] ACPI: AGDI: Add interrupt signaling mode support

[+cc Thomas, Marc]

On Tue, Nov 11, 2025 at 11:42:37PM -0500, Charles Mirabile wrote:
> Hi All—
> 
> On Mon, Nov 10, 2025 at 07:38:17AM +0000, Kazuhiro Abe (Fujitsu) wrote:
> > Hi Will,
> > 
> > > Hi Will,
> > > 
> > > > [You don't often get email from will@...nel.org. Learn why this is
> > > > important at https://aka.ms/LearnAboutSenderIdentification ]
> > > >
> > > > On Mon, Oct 20, 2025 at 09:23:05PM +0800, Hanjun Guo wrote:
> > > > > On 2025/10/17 15:39, Kazuhiro Abe wrote:
> > > > > > AGDI has two types of signaling modes: SDEI and interrupt.
> > > > > > Currently, the AGDI driver only supports SDEI.
> > > > > > Therefore, add support for interrupt signaling mode The interrupt
> > > > > > vector is retrieved from the AGDI table, and call panic function
> > > > > > when an interrupt occurs.
> > > > > >
> > > > > > Reviewed-by: Ilkka Koskinen <ilkka@...amperecomputing.com>
> > > > > > Signed-off-by: Kazuhiro Abe <fj1078ii@...jp.fujitsu.com>
> > > > > > ---
> > > > > > Hanjun, I have addressed all your comments.
> > > > > > Please review them.
> > > > > >
> > > > > > v3->v4
> > > > > >   - Add a comment to the flags member.
> > > > > >   - Fix agdi_interrupt_probe.
> > > > > >   - Fix agdi_interrupt_remove.
> > > > > >   - Add space in struct initializsation.
> > > > > >   - Delete curly braces.
> > > > >
> > > > > Looks good to me,
> > > > >
> > > > > Acked-by: Hanjun Guo <guohanjun@...wei.com>
> > > >
> > > > I wasn't cc'd on the original patch but I couldn't figure out why it
> > > > uses IRQF_NO_AUTOEN when requesting the irq given that the first thing
> > > > it does is enable it.
> > > 
> > > I misunderstood the usage of request_irq and enable_irq.
> > > Since there's no need to separate them, I will remove IRQF_NO_AUTOEN and the
> > > enable_irq call, and send v5.
> > 
> > I found out when calling request_nmi, removing IRQF_NO_AUTOEN results in an error (-EINVAL).
> > Therefore, I would like to keep IRQF_NO_AUTOEN specified.
> > If you have any comments on this version, please let me know.
> 
> Could it be that this is just a bug in `request_nmi`? I see the following:
> 
> if (!desc || (irq_settings_can_autoenable(desc) &&
>     !(irqflags & IRQF_NO_AUTOEN)) ||
>     !irq_settings_can_request(desc) ||
>     WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
>     !irq_supports_nmi(desc))
> 	return -EINVAL;
> 
> Perhaps there is just a missing `!` before `irq_settings_can_autoenable`.
> 
> As far as I can tell it has always been wrong - git blame points me to the
> original commit where that code was introduced:
> 
> b525903c254da ("genirq: Provide basic NMI management for interrupt lines")

I don't think that's the right rationale (and I can't claim to understand
it fully either - that's why I added Thomas/Marc in CC).

As per b525903c254da - IIUC NMI must not be autoenable-capable -
enable/disable must be done using enable_nmi/disable_nmi_nosync explicitly.

cbe16f35bee6 ("genirq: Add IRQF_NO_AUTOEN for request_irq/nmi()")
added the logic checking the IRQF_NO_AUTOEN flag to request_nmi().

Now, AFAICS irq_settings_can_autoenable(desc) returns true in this
specific case - the IRQ is a GICv3 SPI.

First question is - before commit cbe16f35bee6, how request_nmi() could
have possibly succeeded - irq_settings_can_autoenable() would return
true and request_nmi() would have failed.

Come cbe16f35bee6:

if (!desc || (irq_settings_can_autoenable(desc) && !(irqflags & IRQF_NO_AUTOEN)) ||
	!irq_settings_can_request(desc) ||
	WARN_ON(irq_settings_is_per_cpu_devid(desc)) ||
	!irq_supports_nmi(desc))
		return -EINVAL;

Now the check passes _if_ you pass in irqflags IRQF_NO_AUTOEN.

I agree with Will that's a bit counterintuitive - maybe we can force
the irqdesc to be NOAUTOEN in request_nmi() directly ?

Thomas and Marc know better so that's where I stop.

Thanks,
Lorenzo

> I looked and the only two callers are using `IRQF_NO_AUTOEN` so I guess it
> just hasn't been noticed yet.
> 
> Happy to send a patch to fix it.
> 
> > 
> > Best Regards,
> > Kazuhiro Abe
> > 
> > > 
> > > Best Regards,
> > > Kazuhiro Abe
> > > 
> > > >
> > > > Will
> 
> Best—Charlie
>