From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:54114) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XCQ8g-0003KY-Vx for qemu-devel@nongnu.org; Wed, 30 Jul 2014 05:25:45 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1XCQ8a-0004Ew-Rv for qemu-devel@nongnu.org; Wed, 30 Jul 2014 05:25:38 -0400 Received: from mx1.redhat.com ([209.132.183.28]:63621) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1XCQ8a-0004Ej-BB for qemu-devel@nongnu.org; Wed, 30 Jul 2014 05:25:32 -0400 Message-ID: <53D8B9F8.7090804@redhat.com> Date: Wed, 30 Jul 2014 11:25:12 +0200 From: Paolo Bonzini MIME-Version: 1.0 References: <20140717110153.8352.80175.stgit@PASHA-ISP> <53D14707.7080509@redhat.com> <2596.37912172384$1406533875@news.gmane.org> <53D62210.9060903@redhat.com> <000101cfabca$07016870$15043950$@Dovgaluk@ispras.ru> In-Reply-To: <000101cfabca$07016870$15043950$@Dovgaluk@ispras.ru> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [RFC PATCH v2 00/49] Series short description List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Pavel Dovgaluk , qemu-devel@nongnu.org Cc: peter.maydell@linaro.org, peter.crosthwaite@xilinx.com, mark.burton@greensocs.com, real@ispras.ru, batuzovk@ispras.ru, fred.konrad@greensocs.com Il 30/07/2014 09:44, Pavel Dovgaluk ha scritto: >> From: Paolo Bonzini [mailto:paolo.bonzini@gmail.com] On Behalf Of Paolo Bonzini >>>> - patch 16 should also use subsections, and perhaps apply to all other >>>> CPUs too? >>> >>> We implemented replay only for i386 and ARM. If we'll change other targets, it will not >>> add record/replay capabilities to them, but can confuse the reviewers. >> >> Why are these fields only required by record/replay? > > No, these fields can cause non-determinism for normal execution mode too. > But these changes are not related to reverse execution. I can make patches for them > and submit them separately, if needed. Yes, thanks. If it causes nondeterminism for normal execution mode too, it's better to apply them to all targets. If possible use a subsection, though. >>>> - patch 27 makes sense but VIRTUAL is used to skip blinking when the VM >>>> is stopped >>> >>> Right, this is kind of hack. I haven't found better solution yet. >> >> I think it's okay to use REALTIME, just add runstate_is_running() to the >> "if (now >= s->cursor_blink_time)". >> >> That said, "every read to virtual clock is written to the >> replay log" worries me a bit from the point of view of thread-safety. >> Do you need to log all reads because you don't use icount? Reads can >> only happen at given points if you use icount, and you could simply log >> the icount value at each synchronization point. > > We made two implementations of virtual clock. One uses host realtime clock > and thus should be written into the log. Another one uses icount and calculated > in one of the replay modules (similar to regular icount implementation). > This patch series contains only first implementation, because it looks more convenient: > user does not have to specify icount shift value, when starting the recording or replaying. I think from the upstreaming point of view it's better to stick with what makes the code simpler. Start by submitting only the icount-based implementation, the other can come later. Paolo