netdev - Re: [PATCH net-next 1/9] net/mlx4_core: Enforce irq affinity changes immediatly

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]

Message-ID: <1401124850.17377.124.camel@deadeye.wl.decadent.org.uk>
Date:	Mon, 26 May 2014 18:20:50 +0100
From:	Ben Hutchings <ben@...adent.org.uk>
To:	Amir Vadai <amirv@...lanox.com>
Cc:	"David S. Miller" <davem@...emloft.net>, netdev@...r.kernel.org,
	Yevgeny Petrilin <yevgenyp@...lanox.com>,
	Or Gerlitz <ogerlitz@...lanox.com>,
	Yuval Atias <yuvala@...lanox.com>
Subject: Re: [PATCH net-next 1/9] net/mlx4_core: Enforce irq affinity
 changes immediatly

On Mon, 2014-05-19 at 14:33 +0300, Amir Vadai wrote:
> On 5/17/2014 11:33 PM, Ben Hutchings wrote:
> > On Mon, 2014-05-12 at 10:43 +0300, Amir Vadai wrote:
> >> From: Yuval Atias <yuvala@...lanox.com>
> >>
> >> During heavy traffic, napi is constatntly polling the complition queue
> >> and no interrupt is fired. Because of that, changes to irq affinity are
> >> ignored until traffic is stopped and resumed.
> >>
> >> By registering to the irq notifier mechanism, and forcing interrupt when
> >> affinity is changed, irq affinity changes will be immediatly enforced.
> > [...]
> >
> > This somewhat breaks ARFS in your driver, as that depends on IRQ
> > affinity notification.
> Interesting...
> I missed the fact that irq_set_affinity_notifier is 'set' and not 'add'.
> So, do you have objection if I change the API to have a list of 
> notifiers instead of only one?

I don't mind.

> > It is also not safe to put an IRQ affinity notification function in a
> > module currently, as the work item for notification doesn't carry a
> > module reference and could be called after the module is removed.
> I can get a module reference when a notifier is added and put it when 
> the notifier is removed.

Maybe you can drop the reference in the affinity notifier release()
function, though that might still leave a race condition.  But the
module won't be removable until its devices are first removed or at
least brought down (depending on where you add/remove the notifier).

Anyway, I think that the issue you have found is quite general and
should be fixed in the networking core rather than your specific driver.

Ben.

-- 
Ben Hutchings
Humans are not rational beings; they are rationalising beings.

Download attachment "signature.asc" of type "application/pgp-signature" (829 bytes)