The device tree APIC parser tries to force enable the local APIC when it is
not set in CPUID. apic_force_enable() registers the boot CPU apic on
success.
If that succeeds then dtb_lapic_setup() registers the local APIC again
eventually with a different address.
Rewrite the code so that it only registers it once.
Signed-off-by: Thomas Gleixner <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
---
arch/x86/kernel/devicetree.c | 14 +++++---------
1 file changed, 5 insertions(+), 9 deletions(-)
--- a/arch/x86/kernel/devicetree.c
+++ b/arch/x86/kernel/devicetree.c
@@ -157,19 +157,15 @@ static void __init dtb_lapic_setup(void)
/* Did the boot loader setup the local APIC ? */
if (!boot_cpu_has(X86_FEATURE_APIC)) {
+ /* Try force enabling, which registers the APIC address */
if (apic_force_enable(lapic_addr))
return;
- }
- smp_found_config = 1;
- if (of_property_read_bool(dn, "intel,virtual-wire-mode")) {
- pr_info("Virtual Wire compatibility mode.\n");
- pic_mode = 0;
} else {
- pr_info("IMCR and PIC compatibility mode.\n");
- pic_mode = 1;
+ register_lapic_address(lapic_addr);
}
-
- register_lapic_address(lapic_addr);
+ smp_found_config = 1;
+ pic_mode = !of_property_read_bool(dn, "intel,virtual-wire-mode");
+ pr_info("%s compatibility mode.\n", pic_mode ? "IMCR and PIC" : "Virtual Wire");
}
#endif /* CONFIG_X86_LOCAL_APIC */
On 8/1/23 03:46, Thomas Gleixner wrote:
> The device tree APIC parser tries to force enable the local APIC when it is
> not set in CPUID. apic_force_enable() registers the boot CPU apic on
> success.
I went looking for how this registration happens. I was expecting to
see something apic_force_enable() do something similar to
register_lapic_address().
But the apic_force_enable()=>apic_verify() call chain is pretty shallow
and I'm not seeing a lot of overlap.
Am I reading the "registers" part of this too literally?
On Thu, Aug 03 2023 at 15:42, Dave Hansen wrote:
> On 8/1/23 03:46, Thomas Gleixner wrote:
>> The device tree APIC parser tries to force enable the local APIC when it is
>> not set in CPUID. apic_force_enable() registers the boot CPU apic on
>> success.
>
> I went looking for how this registration happens. I was expecting to
> see something apic_force_enable() do something similar to
> register_lapic_address().
>
> But the apic_force_enable()=>apic_verify() call chain is pretty shallow
> and I'm not seeing a lot of overlap.
apic_verify() ends up invoking register_lapic_address() on success, no?