From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ian Campbell Subject: Re: Xen 4.1 rc1 test report Date: Tue, 25 Jan 2011 11:43:18 +0000 Message-ID: <1295955798.14780.5930.camel@zakaz.uk.xensource.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Haitao Shan Cc: "Zheng, Shaohui" , Keir Fraser , "xen-devel@lists.xensource.com" List-Id: xen-devel@lists.xenproject.org On Tue, 2011-01-25 at 06:24 +0000, Haitao Shan wrote: > > Performance(1 bug) > > 1. guest boot very slowly without limit dom0 cpu number on EX (Intel) > > http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1719 > > > > This bug happened 1 year before. Keir has made a fix with c/s 20841, > which essentially holds the locked (and hence allocated) hypercall > page so that next hypercall can reuse it without doing alloc and mlock > again. By doing this, overhead of rschedule IPI as a result of > frequent mlock is greatly reduced. > > Late in year 2010, libxc introduced a new mechanism called hypercall > buffers, as you can refer c/s 22288~22312. Keir's fix is dropped in > this new framework. As a result, the bug appears again. > Probably the new framework auther can pick up Keir's fix again? I think it would make sense to include a low water mark of a small number of pages (perhaps 4 or 8) which instead of being freed are kept and reused in preference to future new allocations. These pages would only finally be released by the xc_interface_close() call. Is this something which you feel able to make a patch for? Ian.