Updated Reports on developer newsgroups suggest that the latest chip stepping of Intel's Itanium processor is causing problems for Linux developers whose code worked fine with stepping A2.
Timing issues seem to be the problem, but it is aggravated by the fact that Intel has so far released few samples of the B1 Itanium, making the process of testing and de-bugging harder than would be the case otherwise.
The Itanium is delayed until the end of this year. Intel officials said at the Developer Forum a few weeks back, that volumes would not begin to ship until next year.
One developer reported: "I got similar results with the 20000501 toolchain come with the 0825 RedHat ia64 beta distribution. I don't think it is a purely compiler issue since the same kernel, which works on A2 stepping, failed to boot on B1 stepping. I suspect some interaction between the ACPI stuff and the interrupts. It seems on B1 stepping, the kernel will get interrupts very early on in the ACPI code."
Another developer put it this way: "In my case, /sbin/hwclock causes the instant reboot on B1 BigSur and it crashes so bad that I have to unplug the power cord to get my machine back. B1 stepping is quite different than A2 stepping on BigSur, even with the same BIOS, Build 59."
But another developer has emailed us to say it is not Intel's fault in the slightest.
Duraid Madina said: "The problem with the Linux kernel has been fixed. Two lines of source in the ACPI (power management) driver need to be changed. The fault lies
entirely with the ACPI driver author, not with Intel!"
And David Mosberger has the final word with this missive: "The Linux kernel works fine on B1 step of Itanium. There was a glitch in the very latest kernel patch which caused a problem on some machines. I was the source of this glitch and am the sole person
responsible for the resulting problems. It has nothing to do with the hardware.
"Please understand that the Linux development happens out in the open, so sometimes a person has a hypothesis that later on turns out to be wrong. In the future, I'd appreciate if you had the courtesy to at least double-check that the main points of your story are true before running it."
We're pleased to hear from Mosberger how these things are thrashed out.
The continuing problems show that entirely re-engineering microprocessors for the IA-64 architecture is far from being a piece of cake. Intel has spent a fortune on pulling in support for ports and encouraging third parties to support its 64-bit effort.
Which just goes to show what a mammoth and Sisyphean task Intel is facing with its behemothic effort. ®