Legacy apic init uses bigsmp for > 8 smp systems.
In these cases, PhysFlat will invariably be used and there
is no point in initializing apic LDR and DFR. Furthermore,
calculate_ldr() helper function was incorrectly setting multiple
bits in the LDR.
This was discovered with a 32 bit KVM guest loading the kdump kernel.
Owing to the multiple bits being incorrectly set in the LDR, KVM hit a
buggy "if" condition in the kvm lapic implementation that creates the
logical destination map for vcpus - it ends up overwriting and
existing valid entry and as a result, apic calibration hangs in the
guest during kdump initialization.
Note that this change isn't intended to workaround the kvm lapic bug;
bigsmp should correctly stay away from initializing LDR.
Suggested-by: Thomas Gleixner <t...@linutronix.de>
Signed-off-by: Bandan Das <b...@redhat.com>
---
arch/x86/kernel/apic/bigsmp_32.c | 24 ++----------------------
1 file changed, 2 insertions(+), 22 deletions(-)
diff --git a/arch/x86/kernel/apic/bigsmp_32.c b/arch/x86/kernel/apic/bigsmp_32.c
index afee386ff711..caedd8d60d36 100644
--- a/arch/x86/kernel/apic/bigsmp_32.c
+++ b/arch/x86/kernel/apic/bigsmp_32.c
@@ -38,32 +38,12 @@ static int bigsmp_early_logical_apicid(int cpu)
return early_per_cpu(x86_cpu_to_apicid, cpu);
}
-static inline unsigned long calculate_ldr(int cpu)
-{
- unsigned long val, id;
-
- val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
- id = per_cpu(x86_bios_cpu_apicid, cpu);
- val |= SET_APIC_LOGICAL_ID(id);
-
- return val;
-}
-
/*
- * Set up the logical destination ID.
- *
- * Intel recommends to set DFR, LDR and TPR before enabling
- * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
- * document number 292116). So here it goes...
+ * bigsmp enables physical destination mode
+ * and doesn't use LDR and DFR
*/
static void bigsmp_init_apic_ldr(void)
{
- unsigned long val;
- int cpu = smp_processor_id();
-
- apic_write(APIC_DFR, APIC_DFR_FLAT);
- val = calculate_ldr(cpu);
- apic_write(APIC_LDR, val);
}
static void bigsmp_setup_apic_routing(void)
--
2.20.1
