From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:50233) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VK1cR-0006Rc-9G for qemu-devel@nongnu.org; Thu, 12 Sep 2013 03:47:21 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1VK1cL-00034e-9n for qemu-devel@nongnu.org; Thu, 12 Sep 2013 03:47:15 -0400 Received: from mx1.redhat.com ([209.132.183.28]:61358) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VK1cK-00034W-VY for qemu-devel@nongnu.org; Thu, 12 Sep 2013 03:47:09 -0400 Message-ID: <5231719B.8000208@redhat.com> Date: Thu, 12 Sep 2013 10:47:39 +0300 From: Orit Wasserman MIME-Version: 1.0 References: <1378285356-5308-1-git-send-email-lilei@linux.vnet.ibm.com> <1378285356-5308-4-git-send-email-lilei@linux.vnet.ibm.com> <877gen3efx.fsf@elfo.elfo> <523038BD.4070706@redhat.com> <8738pb39ed.fsf@elfo.elfo> <523053A8.4000009@redhat.com> In-Reply-To: <523053A8.4000009@redhat.com> Content-Type: text/plain; charset=ISO-8859-15 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 3/3 resend v2] arch_init: right return for ram_save_iterate List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Paolo Bonzini Cc: qemu-devel@nongnu.org, mrhines@linux.vnet.ibm.com, Lei Li , anthony@codemonkey.ws, quintela@redhat.com On 09/11/2013 02:27 PM, Paolo Bonzini wrote: > Il 11/09/2013 13:06, Juan Quintela ha scritto: >>>> And I think that the right solution is make qemu_get_rate_limit() to >>>> return -1 in case of error (or the error, I don't care). >>> >>> You might do both things, it would avoid the useless g_usleep you >>> pointed out below. But Lei's patch is good, because an error could >>> happen exactly during the qemu_put_be64 that writes RAM_SAVE_FLAG_EOS. >> >> Caller checks also. This is the reason I wanted qemu_file_* callers to >> return an error. It has some advantages and some disadvantages. We >> don't agree on which ones are bigger O:-) > > I think the disadvantages are bigger. It litters the code with error > handling, hides where things actually happen, and doesn't even simplify > QEMUFile itself. Checking only at the toplevel is simpler, all we need > to do is ensure that we get there every now and then (and that's what > qemu_file_rate_limit does). > I also prefer the error checking at the top level. Orit >>>> savevm.c: qemu_savevm_state_iterate() >>>> >>>> if (qemu_file_rate_limit(f)) { >>>> return 0; >>>> } >>>> >>>> check is incorrect again, we should return an error if there is one >>>> error. >>> >>> Nothing cares if qemu_savevm_state_iterate returns 0 or negative, so >>> changing qemu_savevm_state_iterate to only return 0/1 would make sense too. >> >> In this case, 0 means: >> please, call us again >> when what we mean is: >> don't care about calling us again, there is an error. Handle the error. > > Or alternatively, 0 means: > > we haven't finished the work > > when what we mean is: > > we haven't finished the work (BTW, please check if there is an error) > >> Notice that qemu_save_iterate() already returns errors in other code >> paths > > Yes that's also unnecessary. > >> If we change th ereturn value for qemu_file_rate_limit() the change that >> cames with this patch is not needed, that was my point. > > This is what an earlier patch from Lei did. I told him (or her?) to > leave qemu_file_rate_limit aside since the idea behind QEMUFile is to > only handle the error at the top. > > Paolo >