Re: Gigabyte 6VXDC7 mobo (Via 694X chipset) SMP instability
Tim Day <[email protected]> Fri, 06 May 2005 08:48:29 +0100
| Newsgroups | org.kernel.vger.linux-smp |
|---|---|
| Message-ID | <[email protected]> |
On Wed, 2005-05-04 at 22:23 -0400, Earle Nietzel wrote: > Tim Day wrote: > > I have a machine with a Gigabyte 6VXDC7 mobo (Via 694X chipset) with a > > pair of 933MHz P3s and a couple of GB of RAM. I run Debian Sarge's > > stock -686-smp kernels (2.4 series or more recently 2.6.8). > > > > With a "nosmp" boot option (or using uniprocessor kernels) the box is > > rock solid. But when I run in SMP mode it generates sporadic > > "APIC error on CPUx: 0y(0z)" > > type messages, generally in pairs, every few minutes (or several per > > minute with the 2.6.8 kernel without a "noapic" boot option). > > Eventually the machine invariably locks up hard with no warning after an > > uptime of on the order of a few minutes or a few days (depending on > > kernel version, boot options, load and luck). I've tried just about > > every possible combination of BIOS settings and boot options, memchecked > > the RAM, rearranged PCI cards and upgraded the PSU. Nothing seems to > > help; the only way to get complete stability seems to be "nosmp" or a > > non-SMP kernel. > > Have you tried using "noapic"? Oh yes. It has the expected effect on /proc/interrupts: IO-APIC-edge/-level interrupts spread over both CPUs change into XT-PIC interrupts on CPU0 only. Nevertheless, the steady stream of "APIC error" kernel messages remains. Is it possible to build an SMP kernel with the APIC hardware completely ignored/disabled (to an even greater extent than with "noapic"), or is the APIC hardware needed for some crucial aspect of inter-CPU communication ? Tim