From mboxrd@z Thu Jan  1 00:00:00 1970
From: Ian Campbell <ian.campbell@citrix.com>
Subject: Re: passthrough: improve interrupt injection locking
Date: Fri, 23 Oct 2015 13:37:25 +0100
Message-ID: <1445603845.2374.140.camel@citrix.com>
References: <1445598322-22154-1-git-send-email-david.vrabel@citrix.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Return-path: <xen-devel-bounces@lists.xen.org>
Received: from mail6.bemta3.messagelabs.com ([195.245.230.39])
	by lists.xen.org with esmtp (Exim 4.72)
	(envelope-from <prvs=7311588c8=Ian.Campbell@citrix.com>)
	id 1Zpbeg-0004Gq-US
	for xen-devel@lists.xenproject.org; Fri, 23 Oct 2015 12:41:11 +0000
In-Reply-To: <1445598322-22154-1-git-send-email-david.vrabel@citrix.com>
List-Unsubscribe: <http://lists.xen.org/cgi-bin/mailman/options/xen-devel>,
	<mailto:xen-devel-request@lists.xen.org?subject=unsubscribe>
List-Post: <mailto:xen-devel@lists.xen.org>
List-Help: <mailto:xen-devel-request@lists.xen.org?subject=help>
List-Subscribe: <http://lists.xen.org/cgi-bin/mailman/listinfo/xen-devel>,
	<mailto:xen-devel-request@lists.xen.org?subject=subscribe>
Sender: xen-devel-bounces@lists.xen.org
Errors-To: xen-devel-bounces@lists.xen.org
To: David Vrabel <david.vrabel@citrix.com>, xen-devel@lists.xenproject.org
Cc: Jan Beulich <jbeulich@suse.com>
List-Id: xen-devel@lists.xenproject.org

On Fri, 2015-10-23 at 12:05 +0100, David Vrabel wrote:
> When injecting an interrupt for a passthrough device into a guest, the
> per-domain event_lock is held, reducing performance when a guest has
> many VCPUs and high interrupt rates.

Did you CC me due to a possible impact on ARM? If so then I think since ARM
lacks this "dpci" stuff none of these changes should have any impact on
that arch.

If you think I've missed something or you CCd me for some other reason
please let me know.

Thanks,
Ian.

> 
> By using a per-interrupt lock in the hot paths, this contention is
> eliminated and performance improves (a bit).
> 
> For testing, a 32 VCPU guest with an NVME device assigned to it was
> used.  Continual reads with small (512 B) blocks were performed on all
> 32 hardware queues simultaneously.
> 
> * Lock profiling:
> 
> Before (elapsed: 60 s):
> 
> (XEN) [ 3321.143155] Domain 1 event_lock:
> (XEN) [ 3321.143158]   lock:    14411627(00000005:90714AEF),
>                       block:     6658599(00000003:709F82BD)
> 
> After (elapsed: 60 s):
> 
> (XEN) [ 1253.921427] Domain 2 event_lock:
> (XEN) [ 1253.921429]   lock:        8287(00000000:01AE517C),
>                       block:          67(00000000:000D4C3A)
> 
> * Aggregate performance:
> 
>         MB/s
> Before  60.8
> After   68.4
> 
> David
>