Re: [PATCH] x86: enable x2apic early at the first point

[Ingo Molnar <mingo@elte.hu>]

* Yinghai Lu <yinghai@kernel.org> wrote:

> Impact: fix bug.
> 
> otherwise will get panic from early_acpi_boot_init()
> also make disable_x2apic global, so could use it it x2_apic_xxx.c
> and can get warning if preenabled system using nox2apic.
> 
> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> 
> ---
>  arch/x86/kernel/apic/apic.c           |    3 +--
>  arch/x86/kernel/apic/x2apic_cluster.c |    5 ++++-
>  arch/x86/kernel/apic/x2apic_phys.c    |    5 ++++-
>  arch/x86/kernel/apic/x2apic_uv_x.c    |    4 +++-
>  drivers/pci/dmar.c                    |    3 ++-
>  5 files changed, 14 insertions(+), 6 deletions(-)

I've applied it because it fixes a real bug, but this code 
really needs a cleanup. Look at the repeat patterns:

> Index: linux-2.6/arch/x86/kernel/apic/x2apic_cluster.c
> ===================================================================
> --- linux-2.6.orig/arch/x86/kernel/apic/x2apic_cluster.c
> +++ linux-2.6/arch/x86/kernel/apic/x2apic_cluster.c
> @@ -14,8 +14,11 @@ DEFINE_PER_CPU(u32, x86_cpu_to_logical_a
>  
>  static int x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
>  {
> -	if (cpu_has_x2apic)
> +	if (cpu_has_x2apic && !disable_x2apic) {
> +		x2apic = 1;
> +		enable_x2apic();
>  		return 1;
> +	}
>  
>  	return 0;
>  }
> Index: linux-2.6/arch/x86/kernel/apic/x2apic_phys.c
> ===================================================================
> --- linux-2.6.orig/arch/x86/kernel/apic/x2apic_phys.c
> +++ linux-2.6/arch/x86/kernel/apic/x2apic_phys.c
> @@ -21,8 +21,11 @@ early_param("x2apic_phys", set_x2apic_ph
>  
>  static int x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
>  {
> -	if (cpu_has_x2apic && x2apic_phys)
> +	if (cpu_has_x2apic && !disable_x2apic && x2apic_phys) {
> +		x2apic = 1;
> +		enable_x2apic();
>  		return 1;
> +	}
>  
>  	return 0;
>  }
> Index: linux-2.6/arch/x86/kernel/apic/x2apic_uv_x.c
> ===================================================================
> --- linux-2.6.orig/arch/x86/kernel/apic/x2apic_uv_x.c
> +++ linux-2.6/arch/x86/kernel/apic/x2apic_uv_x.c
> @@ -41,8 +41,10 @@ static int uv_acpi_madt_oem_check(char *
>  			uv_system_type = UV_LEGACY_APIC;
>  		else if (!strcmp(oem_table_id, "UVX"))
>  			uv_system_type = UV_X2APIC;
> -		else if (!strcmp(oem_table_id, "UVH")) {
> +		else if (!strcmp(oem_table_id, "UVH") && !disable_x2apic) {
>  			uv_system_type = UV_NON_UNIQUE_APIC;
> +			x2apic = 1;
> +			enable_x2apic();
>  			return 1;
>  		}
>  	}

Such repeat patterns with small variations are always the sign 
of an inefficient code structure.

The clean approach would be to have one generic helper function 
concentrated into enable_x2apic(). So instead of:

 static int x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
 {
	if (cpu_has_x2apic && !disable_x2apic && x2apic_phys) {
		x2apic = 1;
		enable_x2apic();
		return 1;
	}

	return 0;
 }

We'd have:

 static int x2apic_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
 {
	if (!x2apic_phys)
		return 0;

	return x86_enable_x2apic();
 }

That way all the repeat and common functionality (and the return 
code logic) is concentrated into enable_x2apic(), and the 
x2apic_acpi_madt_oem_check() function has _only_ its own special 
check open-coded. (namely, whether physical ID delivery mode is 
forced off.)

Okay?

	Ingo