[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20140401093035.GE24150@order.stressinduktion.org>
Date: Tue, 1 Apr 2014 11:30:35 +0200
From: Hannes Frederic Sowa <hannes@...essinduktion.org>
To: ratheesh kannoth <ratheesh.ksz@...il.com>
Cc: netdev <netdev@...r.kernel.org>,
Netfilter mailing list <netfilter@...r.kernel.org>
Subject: Re: cache coherency problem, why not happening ?
On Tue, Apr 01, 2014 at 09:49:53AM +0530, ratheesh kannoth wrote:
> Suppose in smp machine, if bottom half is executed by cpu #1 and
> the userspace process is pinned to cpu #2. How can the spin lock
> acuquired info ( lock value ) will be reflected to cpu #2 ( if cpu
> #2 has already acquired and released the lock value and the lock value
> is stored in a cpu #2 data cache. so if cpu #1 acquire it , cpu#2
> will read the cache value which says it is free). I could see that
> spin_lock_t is not a volatile value also.
volatile is a construct from C-language standard and does not help that much
with syncing to CPUs.
(Specific for Intel processors, now)
If you read the beginning of Chapter 8
<http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-system-programming-manual-325384.pdf>
then lookup ticket spinlocks and then revisit the spin_lock implementation it
should get clear.
Have fun,
Hannes
--
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
Powered by blists - more mailing lists