From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from [140.186.70.92] (port=36590 helo=eggs.gnu.org) by lists.gnu.org with esmtp (Exim 4.43) id 1Pa6fb-00024V-MA for qemu-devel@nongnu.org; Tue, 04 Jan 2011 08:11:25 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Pa6fa-0007Wl-Cm for qemu-devel@nongnu.org; Tue, 04 Jan 2011 08:11:23 -0500 Received: from mx1.redhat.com ([209.132.183.28]:58894) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Pa6fa-0007WJ-1t for qemu-devel@nongnu.org; Tue, 04 Jan 2011 08:11:22 -0500 Date: Tue, 4 Jan 2011 15:10:52 +0200 From: "Michael S. Tsirkin" Subject: Re: [Qemu-devel] Re: [PATCH 09/21] Introduce event-tap. Message-ID: <20110104131052.GA8734@redhat.com> References: <1290665220-26478-1-git-send-email-tamura.yoshiaki@lab.ntt.co.jp> <1290665220-26478-10-git-send-email-tamura.yoshiaki@lab.ntt.co.jp> <20110104111908.GA5694@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline In-Reply-To: Content-Transfer-Encoding: quoted-printable List-Id: qemu-devel.nongnu.org List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Yoshiaki Tamura Cc: aliguori@us.ibm.com, mtosatti@redhat.com, ananth@in.ibm.com, kvm@vger.kernel.org, Stefan Hajnoczi , dlaor@redhat.com, ohmura.kei@lab.ntt.co.jp, qemu-devel@nongnu.org, avi@redhat.com, vatsa@linux.vnet.ibm.com, psuriset@linux.vnet.ibm.com, stefanha@linux.vnet.ibm.com On Tue, Jan 04, 2011 at 09:20:53PM +0900, Yoshiaki Tamura wrote: > 2011/1/4 Michael S. Tsirkin : > > On Tue, Jan 04, 2011 at 08:02:54PM +0900, Yoshiaki Tamura wrote: > >> 2010/11/29 Stefan Hajnoczi : > >> > On Thu, Nov 25, 2010 at 6:06 AM, Yoshiaki Tamura > >> > wrote: > >> >> event-tap controls when to start FT transaction, and provides pro= xy > >> >> functions to called from net/block devices. =A0While FT transacti= on, it > >> >> queues up net/block requests, and flush them when the transaction= gets > >> >> completed. > >> >> > >> >> Signed-off-by: Yoshiaki Tamura > >> >> Signed-off-by: OHMURA Kei > >> >> --- > >> >> =A0Makefile.target | =A0 =A01 + > >> >> =A0block.h =A0 =A0 =A0 =A0 | =A0 =A09 + > >> >> =A0event-tap.c =A0 =A0 | =A0794 +++++++++++++++++++++++++++++++++= ++++++++++++++++++++++ > >> >> =A0event-tap.h =A0 =A0 | =A0 34 +++ > >> >> =A0net.h =A0 =A0 =A0 =A0 =A0 | =A0 =A04 + > >> >> =A0net/queue.c =A0 =A0 | =A0 =A01 + > >> >> =A06 files changed, 843 insertions(+), 0 deletions(-) > >> >> =A0create mode 100644 event-tap.c > >> >> =A0create mode 100644 event-tap.h > >> > > >> > event_tap_state is checked at the beginning of several functions. = =A0If > >> > there is an unexpected state the function silently returns. =A0Sho= uld > >> > these checks really be assert() so there is an abort and backtrace= if > >> > the program ever reaches this state? > >> > > >> >> +typedef struct EventTapBlkReq { > >> >> + =A0 =A0char *device_name; > >> >> + =A0 =A0int num_reqs; > >> >> + =A0 =A0int num_cbs; > >> >> + =A0 =A0bool is_multiwrite; > >> > > >> > Is multiwrite logging necessary? =A0If event tap is called from wi= thin > >> > the block layer then multiwrite is turned into one or more > >> > bdrv_aio_writev() calls. > >> > > >> >> +static void event_tap_replay(void *opaque, int running, int reas= on) > >> >> +{ > >> >> + =A0 =A0EventTapLog *log, *next; > >> >> + > >> >> + =A0 =A0if (!running) { > >> >> + =A0 =A0 =A0 =A0return; > >> >> + =A0 =A0} > >> >> + > >> >> + =A0 =A0if (event_tap_state !=3D EVENT_TAP_LOAD) { > >> >> + =A0 =A0 =A0 =A0return; > >> >> + =A0 =A0} > >> >> + > >> >> + =A0 =A0event_tap_state =3D EVENT_TAP_REPLAY; > >> >> + > >> >> + =A0 =A0QTAILQ_FOREACH(log, &event_list, node) { > >> >> + =A0 =A0 =A0 =A0EventTapBlkReq *blk_req; > >> >> + > >> >> + =A0 =A0 =A0 =A0/* event resume */ > >> >> + =A0 =A0 =A0 =A0switch (log->mode & ~EVENT_TAP_TYPE_MASK) { > >> >> + =A0 =A0 =A0 =A0case EVENT_TAP_NET: > >> >> + =A0 =A0 =A0 =A0 =A0 =A0event_tap_net_flush(&log->net_req); > >> >> + =A0 =A0 =A0 =A0 =A0 =A0break; > >> >> + =A0 =A0 =A0 =A0case EVENT_TAP_BLK: > >> >> + =A0 =A0 =A0 =A0 =A0 =A0blk_req =3D &log->blk_req; > >> >> + =A0 =A0 =A0 =A0 =A0 =A0if ((log->mode & EVENT_TAP_TYPE_MASK) =3D= =3D EVENT_TAP_IOPORT) { > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0switch (log->ioport.index) { > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0case 0: > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0cpu_outb(log->ioport.add= ress, log->ioport.data); > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0break; > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0case 1: > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0cpu_outw(log->ioport.add= ress, log->ioport.data); > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0break; > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0case 2: > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0cpu_outl(log->ioport.add= ress, log->ioport.data); > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0break; > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0} > >> >> + =A0 =A0 =A0 =A0 =A0 =A0} else { > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0/* EVENT_TAP_MMIO */ > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0cpu_physical_memory_rw(log->mmio= .address, > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0= =A0 =A0 =A0 log->mmio.buf, > >> >> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0= =A0 =A0 =A0 log->mmio.len, 1); > >> >> + =A0 =A0 =A0 =A0 =A0 =A0} > >> >> + =A0 =A0 =A0 =A0 =A0 =A0break; > >> > > >> > Why are net tx packets replayed at the net level but blk requests = are > >> > replayed at the pio/mmio level? > >> > > >> > I expected everything to replay either as pio/mmio or as net/block. > >> > >> Stefan, > >> > >> After doing some heavy load tests, I realized that we have to > >> take a hybrid approach to replay for now. =A0This is because when a > >> device moves to the next state (e.g. virtio decreases inuse) is > >> different between net and block. =A0For example, virtio-net > >> decreases inuse upon returning from the net layer, > >> but virtio-blk > >> does that inside of the callback. > > > > For TX, virtio-net calls virtqueue_push from virtio_net_tx_complete. > > For RX, virtio-net calls virtqueue_flush from virtio_net_receive. > > Both are invoked from a callback. > > > >> If we only use pio/mmio > >> replay, even though event-tap tries to replay net requests, some > >> get lost because the state has proceeded already. > > > > It seems that all you need to do to avoid this is to > > delay the callback? >=20 > Yeah, if it's possible. But if you take a look at virtio-net, > you'll see that virtio_push is called immediately after calling > qemu_sendv_packet > while virtio-blk does that in the callback. This is only if the packet was sent immediately. I was referring to the case where the packet is queued. > > > >> This doesn't > >> happen with block, because the state is still old enough to > >> replay. =A0Note that using hybrid approach won't cause duplicated > >> requests on the secondary. > > > > An assumption devices make is that a buffer is unused once > > completion callback was invoked. Does this violate that assumption? >=20 > No, it shouldn't. In case of net with net layer replay, we copy > the content of the requests, and in case of block, because we > haven't called the callback yet, the requests remains fresh. >=20 > Yoshi >=20 Yes, as long as you copy it should be fine. Maybe it's a good idea for event-tap to queue all packets to avoid the copy and avoid the need to replay at the net level. > > > > -- > > MST > > > >