multicast RX broken in joyride

[dsd]

I have an update1 laptop running on mesh network 11 which sees a lot of other XOs in the neighbourhood view.

I then sign my olpc3 laptop onto the same mesh. It appears in the neighbourhood on the update1 system.

No XOs ever appear on the neighbourhood view on the olpc3 system.

[morgs]

I pinged bpepple about building F9 telepathy-gabble/salut packages with olpc patches and --enable-olpc.

[dsd]

confirmed that rebuilding telepathy-salut-0.2.3-1.fc9 with --enable-olpc solves the problem with salut presence.

Can't test gabble presence at the moment because our school server jabber daemon is down

[dsd]

gabble is working

[gdesmott]

So I guess we should close this bug when olpc3 has the right packages.

[dsd]

Agreed. Morgan, has there been any word from Brian?

[dsd]

Morgan says: telepathy-salut with --enable-olpc has been pushed to F9 updates. There are still some patches we need for stream tubes (IIRC) to circumvent normal security in the presence of rainbow that are too crackful to push to F9. dgilmore suggested a runtime option to enable that - cassidy is consulting with daf, sjoerd and m_stone.

[sjoerd]

Replying to dsd:

Morgan says: telepathy-salut with --enable-olpc has been pushed to F9 updates. There are still some patches we need for stream tubes (IIRC) to circumvent normal security in the presence of rainbow that are too crackful to push to F9. dgilmore suggested a runtime option to enable that > - cassidy is consulting with daf, sjoerd and m_stone.

With my salut upstream head on i really want to keep those patches out of salut. They completely circumvent the basic security checking salut and gabble do. If we want to move something like this upstream then it should be fixed in a better way.

[dsd]

Due to time constraints, we have decided to recreate the OLPC-3 branch with our ugly patches. Morgan will take care of this. Hopefully we can find a nicer solution for 9.1. I added this to the nastiness list: http://wiki.laptop.org/go/Distro_Version_Migration_Nastiness

[morgs]

I definitely get salut presence on joyride 2103, with telepathy-salut I built from the OLPC-3 branch.

[morgs]

I have built telepathy-gabble 0.7.4 with OLPC patches on OLPC-3.

Guillaume has built the latest salut 0.3.3 with OLPC patches on OLPC-3 now that we are on F9.

[morgs]

I've built telepathy-gabble 0.7.6 with OLPC patches on OLPC-3.

[morgs]

All these telepathy updates are in joyride as of 2111.

[dsd]

I'm having trouble with joyride-2123. Presence of other XOs only works when associated to a network, no XOs appear when you are on a mesh channel. This is using salut, I'll try gabble when we have a working school server again.

In terms of collaboration, I can't get it working at all. I've been trying to share Chat and Write activities between two XOs over salut (when both on the same access point). When sharing with the neighborhood, the activity appears on the neighborhood view on the other XO but joining it just opens the activity as blank.

I also tried inviting the other XO to the activity, and that didn't result in any visual invitation appearing (but I'm not exactly sure where to be looking for that in this UI redesign).

[morgs]

Please make sure you have logs enabled when testing, and post presenceservice.log, the activity log, and the output of olpc-netstatus - from both XOs

[morgs]

#7457 is probably a duplicate, please coordinate with Charlie.

Sjoerd can possibly help debug this, although he's at Guadec at the moment.

[gdesmott]

Did some tests using Joyride 2155.

Presence and collaboration worked fine with telepathy-salut connected to an AP but not on simple mesh. avahi-browse doesn't show services from the other XO so I suspect a multicast problem in the mesh.

[dsd]

I've been working on this today as well and have made the same findings so far.

[gdesmott]

Seems mdns is completely broken with simple mesh. I published a dummy service using avahi-publish and it was not displayed on my other XO using "avahi-browse --all". avahi-browse still displays local services though and XO's can ping each other.

Re-assigning to the wireless component as I suspect a multicast bug in the mesh code.

[gdesmott]

Replying to dsd:

Can't test gabble presence at the moment because our school server jabber daemon is down

I tested server collaboration with Joyride 2155 and it worked fine.

[dsd]

I have done further testing and I believe joyride cannot receive multicast frames.

I am using this app to test: http://www.venaas.no/multicast/ssmping/

I put 2 708 systems on the same mesh channel. One runs ssmpingd, another runs ssmping, and the ssmping box receives both unicast and multicast responses as expected.

Swap one of the 708 systems for joyride, and repeat the test. Joyride running ssmpingd, 708 running ssmping: 708 receives both unicast and mcast responses - good.

Swap the arrangement so that 708 runs ssmpingd and joyride runs ssmping: joyride only receives unicast responses.

Running libertas firmware 5.110.22p14 on all the systems. This is the version shipped in joyride and in 708.

[cjb]

Try with a stable branch (2.6.22) kernel on the same joyride build?

[dsd]

I determined that joyride-2072 worked and joyride-2081 failed. From that I went through and found that this change in joyride-2074 is what caused the regression:

-libertas-usb8388-firmware.noarch 2:5.110.20.p49-1.fc9
+libertas-usb8388-firmware.noarch 2:5.110.22.p14-1.fc9

Interestingly the version we upgraded to is the same one present in update1. It looks like the joyride kernel is currently incompatible with the firmware being used in update1?

[dsaxena]

Replying to dsd:

I determined that joyride-2072 worked and joyride-2081 failed. From that I went through and found that this change in joyride-2074 is what caused the regression: {{{ -libertas-usb8388-firmware.noarch 2:5.110.20.p49-1.fc9 +libertas-usb8388-firmware.noarch 2:5.110.22.p14-1.fc9 }}} Interestingly the version we upgraded to is the same one present in update1. It looks like the joyride kernel is currently incompatible with the firmware being used in update1?

Probably a regression in the driver brought in from upstream merge. Javier, can you take a look at this one?

[andrey]

Hi. In joyride-2081 it looks like 'depmod' was never run so there is no modules.dep and, therefore, no modules loaded (including libertas).

After running depmod and rebooting, we're fine and avahi shows services advertised on our LAN, so mDNS and multicast are working.

[dsd]

Are you definitely using salut over the mesh?

It works fine for us using salut over infrastructure networks and gabble. It's just the mesh case that is broken.

You could also try the ssmping test.

[andrey]

Ah, sorry, that was (unintentionally) over infra. Turning off multicast on eth0 to get that out of the way:

# ifconfig eth0 -multicast

...and indeed I can't see the other XOs. However it looks like I can't ping via the msh0 interface either, so I'll look into that.

[dsd]

FWIW, unicast pinging over the mesh interface was definitely working here, even when mcast was broken.

[dsd]

The stable kernel branch has several libertas commits which are not in testing. I tried to put them in testing, but it does not seem to have helped the case. My work can be found at users/dsd/olpc-2.6 branch "testing"

[dsd]

Downgrading to 2.6.22 definitely fixes it.

I then tried "backporting" the 2.6.22 libertas driver to 2.6.25. I just copied the directory over verbatim. Booted that kernel, and surprisingly, multicast still does NOT work.

So, this seems to be a kernel regression outside of the wireless driver.

[dsd]

I attempted to reconfirm those results today: I took current testing kernel, copied drivers/net/wireless/libertas from stable, and retested. multicast is working so perhaps I made a mistake yesterday. It looks like something at the driver-level.

Now to resync the drivers... The diff at the moment is:

27 files changed, 1525 insertions(+), 1140 deletions(-)

[dsd]

OK, no idea what went wrong yesterday, but my attempts to forward port the most recent libertas commits from stable to testing do indeed solve the problem.

So, please pull users/dsd/olpc-2.6 branch testing

[dsd]

new kernel pulled in, fixed in joyride-2180

[dsd]

that should say: fixed in joyride-2200

[gdesmott]

Collaboration seems to work fine with Joyride-2216