HyperThreading in recent 2.4-ac kernels

From: Gerald Britton (gbritton@alum.mit.edu)
Date: Wed Dec 11 2002 - 18:30:59 EST

HyperThreading appears to work properly in vanilla 2.4.x, but fails to
initialize the sibling CPUs in 2.4.x-ac. The problem appears to be in
improper indexing by physical vs. logical CPU numbers.

in smpboot.c (in smp_boot_cpus):

        Dprintk("CPU present map: %lx\n", phys_cpu_present_map);
        cpu = 1;
        for (bit = 0; bit < NR_CPUS; bit++) {
                phys_apicid = raw_phys_apicid[bit];
                if ((cpu_to_physical_apicid(bit) == BAD_APICID) &&

in mpparse.c (in MP_processor_info):

        raw_phys_apicid[num_processors - 1] = m->mpc_apicid;

Booting with HT and some debugging enabled yields:

LAPIC (acpi_id[0x0000] id[0x0] enabled[1])
CPU 0 (0x0000) enabledProcessor #0 Pentium 4(tm) XEON(tm) APIC version 16
LAPIC (acpi_id[0x0001] id[0x6] enabled[1])
CPU 1 (0x0600) enabledProcessor #6 Pentium 4(tm) XEON(tm) APIC version 16
LAPIC (acpi_id[0x0002] id[0x1] enabled[1])
CPU 2 (0x0100) enabledProcessor #1 Pentium 4(tm) XEON(tm) APIC version 16
LAPIC (acpi_id[0x0003] id[0x7] enabled[1])
CPU 3 (0x0700) enabledProcessor #7 Pentium 4(tm) XEON(tm) APIC version 16
CPU present map: c3

The processors appear to have physical IDs 0, 1, 6, 7. raw_phys_apicid[] gets
filled at indexes 0-4, but when the kernel tries to boot the CPUs, it queries
it with physical indexes 0, 1, 6, 7 and loses. I'm not sure exactly what the
correct way to fix this is. (a quick hack to raw_phys_apicid does get all 4
CPUs up and apparently working though)

There appear to be other areas where holes in the physical IDs will cause
problems (things fill indexes by a logical cpu number and index later by
physical ID, or the other way around).

Example: following booting a cpu, the check to see if it booted checks
cpu_to_physical_apicid(bit) where bit is the physical cpu id in the map, but
the table it's checking is indexed by logical cpu number.

                                -- Gerald

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

This archive was generated by hypermail 2b29 : Sun Dec 15 2002 - 22:00:23 EST