From mboxrd@z Thu Jan 1 00:00:00 1970 From: Xiao Guangrong Date: Mon, 02 Jul 2012 02:22:08 +0000 Subject: Re: [PATCH 5/6] KVM: Separate rmap_pde from kvm_lpage_info->write_count Message-Id: <4FF105D0.8010400@linux.vnet.ibm.com> List-Id: References: <20120628105733.ccd9abb3.yoshikawa.takuya@oss.ntt.co.jp> <20120628110141.7b908c91.yoshikawa.takuya@oss.ntt.co.jp> <4FEBCBB3.3060707@linux.vnet.ibm.com> <20120628124546.83a829f3.yoshikawa.takuya@oss.ntt.co.jp> <4FEC96EB.1050307@redhat.com> In-Reply-To: <4FEC96EB.1050307@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Avi Kivity Cc: Takuya Yoshikawa , mtosatti@redhat.com, agraf@suse.de, paulus@samba.org, aarcange@redhat.com, kvm@vger.kernel.org, kvm-ppc@vger.kernel.org, linux-kernel@vger.kernel.org, takuya.yoshikawa@gmail.com On 06/29/2012 01:39 AM, Avi Kivity wrote: > But I still think it's the right thing since it simplifies the code. > Maybe we should add a prefetch() on write_count do mitigate the > overhead, if it starts showing up in profiles. > Long time ago, there was a discussion about dropping prefetch in the operation of list walking: http://lwn.net/Articles/444336/ IIRC, the conclusion is that it is better to let CPU prefetch memory by itself. Actually, when i developed lockless spte walking, i measure the performance if prefetch was used, but i did not see the improvement.