| Me: | About Me |
|---|---|
| Things I Have Done: | Linux Toys, Junk |
| Pictures and Stuff: | Planet Horms, Photo Gallery |
| Sites I Maintain: | Verge, Slarken, Ultra Monkey, Super Sparrow |
Once again I've been working away at my patches to move the mapping of EFI on ia64 Xen to match that of ia64 Linux [*].
Why? Because it turns out that you can only teach the firmware about this mapping a maximum of one times between (hard) reboots. This of course is a problem when kexecing between Xen and Linux (aka soft reboot).
Since my last proclamation that things seemed to be working, circa LCA, two fairly serious bugs have been found (by Alex Willamson) and fixed (by me).
The first bug involved registering with the firmware where exactly the handler for PAL and SAL calls exist. I managed to find this problem quite quickly - a stroke of luck to be honest - and it was just a simple matter of teaching the relevant code the new virtual address.
The second bug was that launching an HVM (fully virtualised) domain would cause the hypervisor to lock up - clearly a problem. This took a bit longer to find the cause of and a lot longer to fix. My old enemy TLB repinning was the culprit. But after much pain I seem to have worked out what to do with the VPD (privregs) entry and appear to have a winning combination.
Now that is all finally under control I'm catching up on kexec-tools and LVS patches, some outstanding Debian bugs against packages that I maintain, and finally getting a chance to look at doing something interesting with the xen networking code. Ok, that is a long list, I wonder how far I will get before EFI consumes me again.
Thu, 24 Apr 2008 10:24:47 +1000
These go out to all those who are partial towards either photos of
buildings on Swanson St. near the University of Melbourne or dead trees by
the Yarra River a bit further downstream. Of particular note
was the notice put up by the Bunny
Sherrif nearby some of the dead trees. more...
In unrelated news, my Kexec for ia64 Xen code still appears to be working.
I've upported to the latest ia64 xen-unstable.hg, consolidated some
TLB unpining code, and made a fresh release, v20080206 to celebrate.
Now if only someone other than me would test the code and point
out where the next big bug lies.
more....
Wed, 06 Feb 2008 18:08:19 +0900
I've just got back from a week in Melbourne, attending
Linux.Conf.Au.
Linux.Conf.Au
I'll try and publish some sort of report at some stage, but suffice to say its been an absolute ripper. For those of you who aren't here, or are here but missed something, I encourage you to look at the video's of the presentations that are online, linked from the programme.
Perhaps my favourite was a lightning talk on Greasemonkey by Paul Fenwick. It doesn't seem that the lightening talks were recorded by the AV guys, but an informal summary by Paul is here.
Update: The lightening talks are now online. [OGG] [SPX]. The greasmonkey talk is the very last one for those who feel like seeking. It is also available by iteself on Youtube here.
Getting Home
I flew home via Sydney and the second leg was punctuated by and unexpected stop in Cairns around 2am to fix the de-icing equipment on one of the engines - they said they expected ice further north and couldn't take any risks, fair enough given the layer of snow that was covering Tokyo when I arrived. As the plane was still heavy with fuel, we had to circle around Cairns for an hour or so before landing. In all, we arrived about 2 1/2 hours late. No complains, the staff did a good job keeping everyone informed of what was going on.
The lovely child behind me decided that about an hour or so out of Cairns was a good time to start talking to himself. This got louder and louder and eventually turned into screaming which lasted until breakfast. The poor thing was probably distressed to the max. Though his antics did nothing to soothe myself or nearby passengers.
Just to spice things up an extra bit, the train from the airport into town was delayed - due to snow of course.
Lastly, I've heard stories of people operating checkout-counters paying more attention to the conversation with their colleague than the customer they are serving. But I was surprised to see the same tactic used by the lady who handled paperwork as I prepared to leave Sydney.
Code
On the coding front, I posted an updated version of my kexec for Xen on ia64 patches just before I left Tokyo. I haven't had time to break it yet, but it does solve a bunch of tedious bugs - mainly relating to TLB flushing or pinning - and I'm feeling pretty good about it right now. more...
I have a couple of patches and emails awaiting my attention, mainly relating to LVS, ldirectord or kexec-tools. I will try to attend to them along with my email backlog this week.
Web
Raster and I took a few moments at Linux.Conf.Au to copy the photos that he took at my wedding onto my laptop. I have put an edited version up. more...
Chizuko has also been working on a fun new site, Psychommunity.Org.
Life
Chizuko's visa has finally come through and we are planning to move
ourselves to Sydney in April. So to all our friends down there, see you
soon :-)
Mon, 04 Feb 2008 18:25:23 +0900
Its not often somone interviews me, but this month my number was up. Howard Dahdah of Computerworld spoke to me recently about Linux-HA in the lead up to the Linux-HA BOF at Linux.Conf.au 2008 to be held at Melbourne University later this month / early next month.
Article: The low-down on the Linux High-Availability Project - Cluster improvements deepen Linux HA’s appeal
Update: The article seems to have been syndicated on
Network World,
here.
My number must really be up this time.
Wed, 16 Jan 2008 11:19:44 +0900
As I mentioned in my last
entry for last year I've been working on some tedious problems relating
to re-pining TLB entries when switching region registers on IA64.
The good news is that the fix I found and wrote about really does work. The bad news is that only solves 2 out of 3 cases. VHPT disabled works. VHPT enabled works. per-vcu-VHPT fails.
Using the HW debugger I can look and see what the TLB entries are. Its
rather slow to use, so it isn't practical to check every time an re-pin is
to be done. But spot checks indicate that I'm twiddling the right values.
So I guess its time for some rigorous head-scratching.
Thu, 10 Jan 2008 23:35:33 +0900