From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:48269) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ct7cr-0007eA-UC for qemu-devel@nongnu.org; Wed, 29 Mar 2017 03:02:39 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ct7cn-00006z-V4 for qemu-devel@nongnu.org; Wed, 29 Mar 2017 03:02:38 -0400 Received: from mx1.redhat.com ([209.132.183.28]:53574) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1ct7cn-00006m-Mj for qemu-devel@nongnu.org; Wed, 29 Mar 2017 03:02:33 -0400 Received: from smtp.corp.redhat.com (int-mx04.intmail.prod.int.phx2.redhat.com [10.5.11.14]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 7709C13350 for ; Wed, 29 Mar 2017 07:02:32 +0000 (UTC) Date: Wed, 29 Mar 2017 15:02:25 +0800 From: Peter Xu Message-ID: <20170329070225.GE3152@pxdev.xzpeter.org> References: <20170323204544.12015-1-quintela@redhat.com> <20170323204544.12015-12-quintela@redhat.com> <20170327092350.GH11497@pxdev.xzpeter.org> <8737dxqm1y.fsf@secure.mitica> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <8737dxqm1y.fsf@secure.mitica> Subject: Re: [Qemu-devel] [PATCH 11/51] ram: Move dup_pages into RAMState List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Juan Quintela Cc: qemu-devel@nongnu.org, dgilbert@redhat.com On Tue, Mar 28, 2017 at 08:43:37PM +0200, Juan Quintela wrote: > Peter Xu wrote: > > On Thu, Mar 23, 2017 at 09:45:04PM +0100, Juan Quintela wrote: > >> Once there rename it to its actual meaning, zero_pages. > >> > >> Signed-off-by: Juan Quintela > >> Reviewed-by: Dr. David Alan Gilbert > > > > Reviewed-by: Peter Xu > > > > Will post a question below though (not directly related to this patch > > but context-wide)... > >> { > >> int pages = -1; > >> > >> if (is_zero_range(p, TARGET_PAGE_SIZE)) { > >> - acct_info.dup_pages++; > >> + rs->zero_pages++; > >> *bytes_transferred += save_page_header(f, block, > >> offset | RAM_SAVE_FLAG_COMPRESS); > >> qemu_put_byte(f, 0); > >> @@ -822,11 +826,11 @@ static int ram_save_page(RAMState *rs, MigrationState *ms, QEMUFile *f, > >> if (bytes_xmit > 0) { > >> acct_info.norm_pages++; > >> } else if (bytes_xmit == 0) { > >> - acct_info.dup_pages++; > >> + rs->zero_pages++; > > > > This code path looks suspicous... since iiuc currently it should only > > be triggered by RDMA case, and I believe here qemu_rdma_save_page() > > should have met something wrong (so that it didn't return with > > RAM_SAVE_CONTROL_DELAYED). Then is it correct we do increase zero page > > counting unconditionally here? (hmm, the default bytes_xmit is zero as > > well...) > > My head hurts at this point. Sorry about that! :( > ok. bytse_xmit can only be zero if we called qemu_rdma_save_page() with > size=0 or there has been an RDMA error. We ver call the function with > size = 0. And if there is one error, we are in very bady shape already. > > > Another thing is that I see when RDMA is enabled we are updating > > accounting info with acct_update_position(), while we updated it here > > as well. Is this an issue of duplicated accounting? > > I think stats and rdma are not right. I have to check more that. Sorry to have led the discussion too far away from the topic. I guess it'll be perfectly okay to just mark this as TODO item, and we can just move on with current series (and I believe you have further patches after this big one :). Out of curiosity - to what extent are people using migration with RDMA? Should that be "very rare"? Thanks, -- peterx