[Cerowrt-devel] 10 steps back, a few steps forward

[Dave Taht]

You can only run at 104% for so long. We really hit a wall in the last
few weeks,
some nagging problems became showstoppers, with some surprising
failures elsewhere...

and I had other huge demands on my time, like trying to find a new
test platform for ath10k gear... and writing/sheparding 5 internet
drafts into the ietf by feb 14th... (One of the five landed in time, 3
of the remaining are very close to completion (actually one landed *3*
minutes too late for the cutoff). Even though it's past the deadline,
getting them done soon would be a relief, even if they have to wait
til next ietf to be presented)

I haven't got much of anything done on my day job in a month.

A change to openwrt's build process (mostly relating to busybox) broke
the cerowrt build
a few weeks ago. Lacked braincells to figure it out.

A critical server crashed over christmas. Build quality from the
openwrt build cluster declined considerably.

And my car got broken into while I was transporting the lab somewhere
warmer and less wet last week.

Anyway, some forward progress to report

+ I THINK the last source specific routing crash bug is licked (thx
Hans Dedecker)
+ some replacements for the stolen test hardware are arriving
+ the dead build server has been fixed and will go back online next week
   ... and hopefully we've lined up a few other parties to help on
openwrt build server front
+ A replacement for the mirror server is nearly done, but isn't racked
yet, and should go up the  next time I am in the co-lo
+ Rebuild cero from cleaner sources... mostly done. A busted build is
replicable at least. :/
+ Sebastian has continued work on the SQM system
+ the dnssec work goes well (still needs nsec3 I guess)
+ the source specific routing + dns problem's solution has been
discussed with simon, I
   don't know if the openwrt devs bought into the simplest solutions
+ The new taller nuc is working out ok as a headless test box with a
normal 2.5in
    sata drive and a full length ath10k card. I have some
"interesting" results from
    adding a usb3 ethernet to it as a second port, which I guess I
should  publish...

   (Still am looking for a
   decent dual-ethernet board (waiting on toke's eval of his choice)
with mini-pci slots.)

+ It looks like we can finally get away from using udev (I could be
wrong though (?))

+/- Avahi is the only thing that pulls in dbus. The plan has long been
to replace that
   with the hybrid mdns proxy work which has stalled out. Getting rid
of dbus and avahi
   saves 1.5MB or so, and hopefully fix the nagging problems with .local naming.

+/- at the moment it looks like openwrt is going to stick with 3.10
for the stable
Barrier Breaker release on this chipset, due in a quarter or two. This
is a plus (it's stable), and a minus (3.13 support has landed for most
other arches). I have no problem with moving to 3.13 given the
schedules I've heard, it's not my call.

- The unaligned instruction traps are baaaaack with a vengance. On my last build
I hit 64000 of them before I tried to look at the cause and locked up the box.
(box had something like 8 ipv6 addresses on it, but they didn't seem
to be related.
 toolchain change from last build? uncommitted or mangled patch? don't know)

- Pim, already broken, broke more.
(this is also a plus, this was apparently an uncommited patch of mine)
- haven't updated to the final pie code
- I'm too chicken to put in dnsmasq-dnssec yet
- no babeld procd support
- am confused about upnp and minidssd
- wanted to take a hard look at the new hncp protocol

I'd like to get to a stabler release as much as every one here,
probably more. :/

I am increasingly uncertain it will be this month. I have talks to
prepare for queen mary and cambridge university that are unwritten,
and other stuff to do to get ready for on this trip, and a backlog of
day-job work...

My goal today is to get a usable development build to be able to attack the
!@#! unaligned problem some more. And then stop.

OK, it's off to make the donuts, this morning.

PS Anyway, am really happy this was found and fixed, the effects were quite
nasty:

commit fc80c8d40f5459b4314c8e70d8fbd9b99ed7f8a1
Author: cyrus <cyrus at 3c298f89-4303-0410-b956-a3cf2f4a3e73>
Date:   Fri Feb 14 21:21:44 2014 +0000

    netifd: fix source-routing breaking 6in4 + others

    6in4 historically allowed an ip6addr without a mask however the newly
    introduced source-routing segfaulted in this scenario (#14958 + #14858).

    Fixes include: "Fix ubus route src mask printing" and
    "Disable netlink auto ack" (thanks to Hans Dedecker)


-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html