On Fri, Jul 20, 2012 at 12:18:59AM +0300, Gleb Natapov wrote:
On Thu, Jul 19, 2012 at 05:52:41PM -0300, Eduardo Habkost wrote:
When resetting an I/O APIC, its ID is set to 0, so set it to 0 on the MADT table too.
Actually BIOS needs to configure ioapic id to a uniqe value. This does not really matter for KVM though.
Where does this requirement comes from? I am guessing it matters only when the I/O APIC is directly connected to the APIC bus (according to Intel SDM, that's the case only for old Pentium and P6 CPUs)[1].
Anyway, even if some hardware has this unique-ID requirement, today Seabios does not fulfill it, leaving the I/O APIC ID as 0. The patch at least makes the MADT table match reality.
[1] I have checked 3 different machines, and all machines I have looked have an I/O APIC ID that conflicts with an existing Local APIC ID, on the ACPI MADT table.
Some iasl dumps may be found online by googling for: "Subtable Type : 01" "I/O Apic ID"
I looked at 5 or 6 matches, and almost every one have an I/O APIC ID conflicting with a Local APIC ID.
Signed-off-by: Eduardo Habkost ehabkost@redhat.com
src/acpi.c | 2 +- src/config.h | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-)
diff --git a/src/acpi.c b/src/acpi.c index 55e4607..3f55de9 100644 --- a/src/acpi.c +++ b/src/acpi.c @@ -335,7 +335,7 @@ build_madt(void) struct madt_io_apic *io_apic = (void*)apic; io_apic->type = APIC_IO; io_apic->length = sizeof(*io_apic);
- io_apic->io_apic_id = CountCPUs;
- io_apic->io_apic_id = BUILD_IOAPIC_ID; io_apic->address = cpu_to_le32(BUILD_IOAPIC_ADDR); io_apic->interrupt = cpu_to_le32(0);
diff --git a/src/config.h b/src/config.h index 3a70867..878c691 100644 --- a/src/config.h +++ b/src/config.h @@ -52,9 +52,11 @@ #define BUILD_PCIMEM64_END 0x10000000000ULL
#define BUILD_IOAPIC_ADDR 0xfec00000 +#define BUILD_IOAPIC_ID 0 #define BUILD_HPET_ADDRESS 0xfed00000 #define BUILD_APIC_ADDR 0xfee00000
// Important real-mode segments #define SEG_IVT 0x0000
#define SEG_BDA 0x0040
1.7.10.4
-- Gleb.