From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeremy Fitzhardinge Subject: Re: boot_cpu_data differs between Xen hypervisor boot and normal pv_ops kernel boot Date: Wed, 31 Mar 2010 11:14:10 -0700 Message-ID: <4BB390F2.8080206@goop.org> References: <4BB36A40.1080903@purdue.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <4BB36A40.1080903@purdue.edu> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Naresh Rapolu Cc: xen-devel@lists.xensource.com, Stephen Spector List-Id: xen-devel@lists.xenproject.org On 03/31/2010 08:29 AM, Naresh Rapolu wrote: > > When I looked into the source code, > arch/x86/kernel/cpu/perf_event.c, at run-time boot_cpu_data > structure seems to differ in these two forms of booting. > Due to this, Iam not able to use the hardware performance counters in > my Xeon 5530 processor as hardware events(cache-misses etc) are > being disabled. > > Is there any reason for this ? The PMU hardware features are not virtualized, and are not available to guest domains. I don't think Xen has much of a useful API to expose these features, though oprofile was working at one point. I agree it would be very useful to work out how to get good "perf" support in a Xen domain. (Stephen: It might be an interesting SoC project.) J