From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from metis.ext.pengutronix.de ([2001:67c:670:201:290:27ff:fe1d:cc33]) by bombadil.infradead.org with esmtps (Exim 4.92.3 #3 (Red Hat Linux)) id 1jbJrq-0008Mt-6a for barebox@lists.infradead.org; Wed, 20 May 2020 08:14:25 +0000 Date: Wed, 20 May 2020 10:14:20 +0200 From: Sascha Hauer Message-ID: <20200520081420.GT11869@pengutronix.de> References: <20200520055231.GR11869@pengutronix.de> <20200520065744.GA14518@homes.emlix.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <20200520065744.GA14518@homes.emlix.com> List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="iso-8859-15" Content-Transfer-Encoding: quoted-printable Sender: "barebox" Errors-To: barebox-bounces+u.kleine-koenig=pengutronix.de@lists.infradead.org Subject: Re: [PATCH v2 09/10] fastboot net: implement fastboot over UDP To: Daniel =?iso-8859-15?Q?Gl=F6ckner?= Cc: barebox@lists.infradead.org, Edmund Henniges Hi Daniel, On Wed, May 20, 2020 at 08:57:44AM +0200, Daniel Gl=F6ckner wrote: > Hello Sascha, > = > On Wed, May 20, 2020 at 07:52:32AM +0200, Sascha Hauer wrote: > > On Thu, May 14, 2020 at 08:21:57PM +0200, Daniel Gl=F6ckner wrote: > > > From: Edmund Henniges > > > = > > > This implements the UDP variant of the fastboot protocol. The only wa= y to > > > start the service for now is to compile with CONFIG_FASTBOOT_NET_ON_B= OOT. > > > The service will bind to the network interface that provides the IPv4 > > > gateway. > > > = > > > Sending an OKAY packet before performing a restart is necessary since > > > contrary to USB the host will not notice when a UDP server disappears. > > > = > > > Signed-off-by: Edmund Henniges > > > Signed-off-by: Daniel Gl=F6ckner > > > --- > > > common/fastboot.c | 3 + > > > include/fastboot.h | 1 + > > > include/fastboot_net.h | 12 + > > > net/Kconfig | 18 ++ > > > net/Makefile | 1 + > > > net/fastboot.c | 496 +++++++++++++++++++++++++++++++++++++++= ++ > > > 6 files changed, 531 insertions(+) > > > create mode 100644 include/fastboot_net.h > > > create mode 100644 net/fastboot.c > > > = > > > diff --git a/common/fastboot.c b/common/fastboot.c > > > index d50f61ff2..706547309 100644 > > > --- a/common/fastboot.c > > > +++ b/common/fastboot.c > > > +static int fastboot_write_net(struct fastboot *fb, const char *buf, > > > + unsigned int n) > > > +{ > > > + struct fastboot_net *fbn =3D container_of(fb, struct fastboot_net, > > > + fastboot); > > > + struct fastboot_header response_header; > > > + uchar *packet; > > > + uchar *packet_base; > > > + > > > + if (fbn->reinit) > > > + return 0; > > > + > > > + while (fbn->may_send =3D=3D MAY_NOT_SEND) { > > > + poller_call(); > > > + if (fbn->reinit) > > > + return 0; > > > + } > > = > > This is a potentially endless loop I can reproducibly hit here. When I > > do a "sleep 10" on the command line and at the same time start a > > fastboot transfer then there's no progress. This is expected because > > fastboot depends on the idle slice. However, when the "sleep 10" is over > > then I get a prompt and then we are stuck in this loop. > > = > > I replaced this loop with a one second timeout loop. With this we are no > > longer stuck in the loop, but the current transfer doesn't continue > > anymore. > = > The official fastboot tool from Google has a 60 second timeout and > retransmits packets every 0.5 seconds. A 1 second timeout on the device > feels too short. We should also retry for at least 60 seconds. > = > I would expect that the host retransmits its last packet beyond the 10 > second sleep and the transfer to immediately continue when barebox is > no longer busy. Also keep in mind that a device will most likely never > be used over fastboot and serial console at the same time. A user who > interactively started the transfer can always restart fastboot on the > host, which will abort this loop by setting reinit. > = > > Afterwards I dropped everything around fbn->may_send completely and now > > everything works fluently. > > = > > Removing these send restrictions seems the right thing to me. In the end > > it's UDP, so the remote host shouldn't make any assumptions of how fast > > and if at all packets arrive there. > = > I haven't seen your changes, but I fear you have broken something else. > Without may_send it may be possible for a message to be sent before the > previous message has been acked by the host. Since we can retransmit > only the last message, the connection will be stuck until the host times > out. We are probably implementing UDP Protocol V1, right? Here it says: 1. The host drives all communication; the device may only send a packet as a response to a host packet. 2. If the host does not receive a response in 500ms it will re-transmit. This means the host doesn't ack anything, it just sends the next packet if it received our response or the same packet again if it hasn't. For us this means we don't have to retransmit anything, we only have to respond to the last packet seen. I am not sure how INFO responses are handled, there seems to be no way for the client to notice that the host hasn't seen them. It looks like these are just lost then and simply won't be displayed on the host. That said, it's absolutely possible that I have broken something else with my changes ;) I have quite some changes also in my slice series which I'll post later the day. Maybe we can take this as a base for further discussions. Sascha -- = Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | _______________________________________________ barebox mailing list barebox@lists.infradead.org http://lists.infradead.org/mailman/listinfo/barebox