netdev - Re: [PATCH net] bna: page allocation during interrupts to use a mempool.

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]

Message-ID: <1412654692.11091.105.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Mon, 06 Oct 2014 21:04:52 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Eric Wheeler <netdev@...ts.ewheeler.net>
Cc:	Shahed Shaikh <shahed.shaikh@...gic.com>,
	Stephen Hemminger <stephen@...workplumber.org>,
	netdev <netdev@...r.kernel.org>,
	Rasesh Mody <rasesh.mody@...gic.com>
Subject: Re: [PATCH net] bna: page allocation during interrupts to use a
 mempool.

On Mon, 2014-10-06 at 20:05 -0700, Eric Wheeler wrote:

> The patch in question implements a simple mempool for the interrupt page 
> allocs---which definitely fixes the problem even if a better solution 
> might exist.  I have no problem giving up a small amount of memory to 
> guarantee page allocs in the interrupt handler.

You have no such guarantee.

Once page is consumed by some networking frame, this frame can be
sitting in some socket receive queue. page cant be reused by the driver.

If you receive (small) burst of 100 frames, your 32 mempool will be
depleted anyway.

Even if you use a mempool with 10000 elements, there is no guarantee, it
really depends on the number of sockets and their SO_RCVBUF limits.

If your NIC depends of having order-2 pages available for its operation,
then I am afraid it cannot possibly work.

Memory will be eventually fragmented and order-2 allocations fail.

Run your patch with 100 concurrent TCP flows, add some losses to force
usage of out or order queue, you'll get errors quite fast.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html