Marcelo, All,
I noticed on x440s that when HT is disabled in the BIOS I was having
problems properly booting 2.4 in ACPI mode. Further investigation found
a subtle problem w/ smp_boot_cpus() when clustered_acpi_mode is set.
During bootup, phys_cpu_present_map is initialized by ORing
apicid_to_phys_cpu_present() for each cpu apicid(see MP_processor_info)
On flat mode boxes this translates to "phys_cpu_present_map |=
(1<<apicid)".
On clustered_apic_mode boxes, since we're using phyiscal apic addresses,
the apicids are not sequential so it is possible the
phys_cpu_present_map can have holes in it (see
apicid_to_phys_cpu_present()).
The problem arises in smp_boot_cpus() because when we are booting the
cpus, we iterate through each apicid, however we bit-AND
phys_cpu_present_map w/ (1<<apicid) rather then using
apicid_to_phys_cpu_present(apicid). On clustered apic machines, this may
cause us to try to boot apicids that do not exist.
The following patch corrects the problem by always bit-ANDing
phys_cpu_present_map with apicid_to_phys_cpu_present(). This is safe for
flat mode boxes, as apicid_to_phys_cpu_present(apicid) translates back
to (1<<apicid).
Additionally, the patch insures we do not try to boot BAD_APICIDs and
removes a hack that was added to mpparse.c which worked around this
problem in the non-ACPI boot path.
In 2.5 we do not have this problem as we use logical rather then
physical apic addressing.
Please consider for inclusion in your tree.
thanks
-john
diff -Nru a/arch/i386/kernel/mpparse.c b/arch/i386/kernel/mpparse.c
--- a/arch/i386/kernel/mpparse.c Mon Nov 3 13:48:33 2003
+++ b/arch/i386/kernel/mpparse.c Mon Nov 3 13:48:33 2003
@@ -587,10 +587,6 @@
++mpc_record;
}
- if (clustered_apic_mode){
- phys_cpu_present_map = logical_cpu_present_map;
- }
-
printk("Enabling APIC mode: ");
if(clustered_apic_mode == CLUSTERED_APIC_NUMAQ)
diff -Nru a/arch/i386/kernel/process.c b/arch/i386/kernel/process.c
--- a/arch/i386/kernel/process.c Mon Nov 3 13:48:33 2003
+++ b/arch/i386/kernel/process.c Mon Nov 3 13:48:33 2003
@@ -44,6 +44,7 @@
#include <asm/irq.h>
#include <asm/desc.h>
#include <asm/mmu_context.h>
+#include <asm/smpboot.h>
#ifdef CONFIG_MATH_EMULATION
#include <asm/math_emu.h>
#endif
@@ -377,7 +378,7 @@
if its not, default to the BSP */
if ((reboot_cpu == -1) ||
(reboot_cpu > (NR_CPUS -1)) ||
- !(phys_cpu_present_map & (1<<cpuid)))
+ !(phys_cpu_present_map & apicid_to_phys_cpu_present(cpuid)))
reboot_cpu = boot_cpu_physical_apicid;
reboot_smp = 0; /* use this as a flag to only go through this once*/
diff -Nru a/arch/i386/kernel/smpboot.c b/arch/i386/kernel/smpboot.c
--- a/arch/i386/kernel/smpboot.c Mon Nov 3 13:48:33 2003
+++ b/arch/i386/kernel/smpboot.c Mon Nov 3 13:48:33 2003
@@ -1108,13 +1108,17 @@
for (bit = 0; bit < NR_CPUS; bit++) {
apicid = cpu_present_to_apicid(bit);
+
+ /* don't try to boot BAD_APICID */
+ if (apicid == BAD_APICID)
+ continue;
/*
* Don't even attempt to start the boot CPU!
*/
if (apicid == boot_cpu_apicid)
continue;
- if (!(phys_cpu_present_map & (1ul << bit)))
+ if (!(phys_cpu_present_map & apicid_to_phys_cpu_present(apicid)))
continue;
if (max_cpus <= cpucount+1)
continue;
@@ -1125,7 +1129,8 @@
* Make sure we unmap all failed CPUs
*/
if ((boot_apicid_to_cpu(apicid) == -1) &&
- (phys_cpu_present_map & (1ul << bit)))
+ (phys_cpu_present_map &
+ apicid_to_phys_cpu_present(apicid)))
printk("CPU #%d/0x%02x not responding - cannot use it.\n",
bit, apicid);
}