linux-kernel - Re: [PATCH v2 1/3] genirq/affinity: Add irq_update_affinity

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]

Date:   Thu, 19 Nov 2020 09:31:56 +0000
From:   John Garry <john.garry@...wei.com>
To:     Thomas Gleixner <tglx@...utronix.de>, <gregkh@...uxfoundation.org>,
        <rafael@...nel.org>, <martin.petersen@...cle.com>,
        <jejb@...ux.ibm.com>
CC:     <linuxarm@...wei.com>, <linux-scsi@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>, <maz@...nel.org>
Subject: Re: [PATCH v2 1/3] genirq/affinity: Add irq_update_affinity_desc()

Hi Thomas,

>>> +int irq_update_affinity_desc(unsigned int irq,
>>> +			     struct irq_affinity_desc *affinity)
>> Just a note on the return value, in the only current callsite -
>> platform_get_irqs_affinity() - we don't check the return value and
>> propagate the error. This is because we don't want to fail the interrupt
>> init just because of problems updating the affinity mask. So I could
>> print a message to inform the user of error (at the callsite).
> Well, not sure about that. During init on a platform which does not have
> the issues with reservation mode there failure cases are:
> 
>   1) Interrupt does not exist. Definitely a full fail
> 
>   2) Interrupt is already started up. Not a good idea on init() and
>      a clear fail.
> 
>   3) Interrupt has already been switched to managed. Double init is not
>      really a good sign either.

I just tested that and case 3) would be a problem. I don't see us 
clearing the managed flag when free'ing the interrupt. So with 
CONFIG_DEBUG_TEST_DRIVER_REMOVE=y, we attempt this affinity update 
twice, and error from the irqd_affinity_is_managed() check.

> 
>>> +	/* Requires the interrupt to be shut down */
>>> +	if (irqd_is_started(&desc->irq_data))
>> We're missing the unlock here, right?
> Duh yes.
> 
>>> +		return -EBUSY;
>>> +
>>> +	/* Interrupts which are already managed cannot be modified */
>>> +	if (irqd_is_managed(&desc->irq_data))
>> And here, and I figure that this should be irqd_affinity_is_managed()
> More duh:)
> 
> I assume you send a fixed variant of this.

Can do.

Thanks,
John