09/11/06 – took a day off to run necessary errands, but managed to spend a couple of hours at night with the CalIOC2 machine in Beaverton. CalIOC2 is a PCI-e bridge. It looks like the adapters behind the CalIOC2 are actually behind a PCI-to-PCI bridge, which is something the Calgary code doesn’t know how to handle. By the time I’m done with it, it will 🙂
P.S. This is amazing.
07/11/06 – more debugging of Xen in an attempt to perform an elusive and highly non trivial, in fact, downright *unusual*, hypervisor task
– creating a new domain. Eventually, learned to ignore the misleading
tool messages, got the versions of kernel, hypervisor and tools in sync and found out the kernel I was trying to boot can’t work as a domU kernel. D’oh! at this point it was midnight and the baby had worn me out completely so went to sleep.
08/11/06 – woke up and found an email with the calgary-recognize-calioc2 boot log. It wasn’t pretty, we were dying with a NULL pointer dereference in spin_lock_irqsave because of a NULL iommu_table. Quick perusal of the code revealed that I’m a doofus and only handled CalIOC2 in one place rather than the requisite two. Rewrote the patch – leading to a couple of nice cleanups in the old Calgary code – and shipped it off to be tested again.
05/11/06 – Spotted an annoying thinko in the calgary-bios-cleanup patch and sent Andi Kleen a fix. Also looked into hardware for next year’s planned escapades, listened to talk by Stu Feldman (currently IBM VP, but previously the guy who invented make) on Exploratory Research and Services Science, and installed an ssh client on the mobile toy. Hee hee!
06/11/06 – Spent a few frustrating hours trying to get xen-unstable to create a new domain, so that I could give it direct access to a NIC. Failed miserably. Not quite sure what’s wrong yet, except that do_domctl() is returning -1. Probably an incompatibility between the iommu-xenlinux tree and the xen-unstable tools, but more head banging required. Useful tip – always compile the tools with XEN_PYTHON_NATIVE_INSTALL=y if you have more than one python version installed. Also worked on a research proposal for next year on IOMMU performance and listened to Eliot Jaffe’s talk about the Internet Archive’s storage woes.
01/11/06 – on the Linux front, worked on two patches: a cleanup patch for Laurent’s excellent calgary-locate-bbars patch and a patch to recognize CalIOC2, the follow on chip to Calgary, which is mostly compatible. Other than that, the Other Project kept me busy.
02/11/06 – Messed with git and hg. I’ve given up on `hg email’ for sending kernel patches, but git-send-email works just fine once you get past the idiosyncrasies. Submitted the latest Calgary patchset to Andi Kleen, who promptly went and applied it to his tree. Woot 🙂 It includes the calgary-locate-bbars patches and a patch to make it possible to compile Calgary in but not use it by default, which is mostly useful for the distros.
03/11/06 – Weekend! didn’t get any work done, unless writing this counts as work 🙂
Highlight of the day: Calendar updates from Lotus Notes are now automatically sent – via procmail – to my mobile toy^H^H^Hphone.
I spent most of the day today working on the Other Project. On the Linux/Xen front I partook in the usual merging activity, as well as testing djwong’s aic94xx patches and reviewing Laurent’s calgary-locate-bbar patch. I’m very happy with this patch, it allows us to get rid of an awful hack – going from dev->bus->number to RIO node ID and BBAR address. It’s awful because it relies on dev->bus->number and the formula being used remaining stable across kernel, BIOS and chipset revisions, which is highly unlikely. In fact, what prompted Laurent’s patch was the forumla being wrong for an x460 configured in a certain way. With the patch we now get the right values for each PHB from the BIOS supplied tables.
I decided to experiment with the “activity log” format again.
30/10/06 – Spent the day at the Intel Developer Forum in Tel-Aviv. Interesting tidbits: VT-d is going to be shipping in ’07, next generation VT-x will have Extended Page Tables and a tagged TLB. Basically guest page tables will translate from virtual to guest physical and EPT will translate from guest physical to machine addresses. The TLB will include the entire translation (hence the need for a tagged TLB to avoid a TLB flush when doing a VM switch). Didn’t get any hacking done but talked to lots of interesting people.
29/10/06 – Linux: wrote patch for mainline to make it possible to compile Calgary in but not use it (patch is at http://www.kernel.org/pub/linux/kernel/people/muli/calgary/calgary-choose-to-use). This is mostly for the distros. Xen: got stock xenlinux running on undeg with aic94xx, verified 3com NIC works fine in dom0.
28/10/06 – Debugging Laurent’s Calgary bug (http://bugzilla.kernel.org/show_bug.cgi?id=7407). Calgary’s BBAR register was getting initialized by the BIOS to a different address than the one we were expecting. Laurent put together a preliminary patch to get the BBAR address from the BIOS supplied tables rather than calculating it directly. This is somewhat problematic as the necessary headers are only defined for i386, whereas Calgary lives in arch/x86_64, so we’ll need to figure out a clean way to use them.
Yael Ben-Yehuda was born at 2:35 PM yesterday, weighing 3.415 kgs. Mother, baby and father are all doing great. I’m sure ladypine will have more to say once she recuperates.
I’ll be mostly offline for the next few days, for obvious reasons.
Muli “Who’s Your Daddy” Ben-Yehuda.
Muli Ben-Yehuda's journal 