netdev - Re: [PATCH net-next v4 5/7] eventpoll: Control irq suspension for prefer_busy

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: <Zyb3yS6Whi5Na0lg@LQ3V64L9R2>
Date: Sat, 2 Nov 2024 21:10:49 -0700
From: Joe Damato <jdamato@...tly.com>
To: Hillf Danton <hdanton@...a.com>
Cc: netdev@...r.kernel.org, edumazet@...gle.com, willy@...radead.org,
	Alexander Viro <viro@...iv.linux.org.uk>,
	linux-kernel@...r.kernel.org, mkarsten@...terloo.ca,
	sridhar.samudrala@...el.com
Subject: Re: [PATCH net-next v4 5/7] eventpoll: Control irq suspension for
 prefer_busy_poll

On Sun, Nov 03, 2024 at 07:39:25AM +0800, Hillf Danton wrote:
> On Sat,  2 Nov 2024 00:52:01 +0000 Martin Karsten <mkarsten@...terloo.ca>
> > 
> > @@ -2005,8 +2032,10 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
> >  			 * trying again in search of more luck.
> >  			 */
> >  			res = ep_send_events(ep, events, maxevents);
> > -			if (res)
> > +			if (res) {
> > +				ep_suspend_napi_irqs(ep);
> 
> Leave napi irq intact in case of -EINTR.

(I've added Martin and Sridhar to the CC list)

Thanks for pointing out this inconsistency. It's not a big problem,
because on receiving EINTR or another error code, the app either
retries epoll_wait or, if the app is buggy somehow and doesn't retry
epoll_wait, the timer fires and everything proceeds as normal.

However, since irqs are not suspended at other points in ep_poll
where an error code is returned, it is probably best to be
consistent and not suspend here either. We will fix this in the next
revision.

Sridhar: Since the change is very minor I plan to retain your
Reviewed-by, but if you'd like me to drop it, please let me know.

The proposed fix will look like:

if (res) {
  if (res > 0)
    ep_suspend_napi_irqs(ep);
  return res;
}

> 
> >  				return res;
> > +			}
> >  		}
> >  
> >  		if (timed_out)
> > -- 
> > 2.25.1