Re: [PULL -tip] x86: replacing mp_config_X with mpc_X

[Jaswinder Singh Rajput <jaswinderlinux@gmail.com>]

On Mon, Jan 12, 2009 at 6:41 PM, Ingo Molnar <mingo@elte.hu> wrote:
>
> * Jaswinder Singh Rajput <jaswinder@infradead.org> wrote:
>
>> -struct mp_config_ioapic {
>> -     unsigned long mp_apicaddr;
>> -     unsigned int mp_apicid;
>> -     unsigned char mp_type;
>> -     unsigned char mp_apicver;
>> -     unsigned char mp_flags;
>> -};
>> -
>> -struct mp_config_intsrc {
>> -     unsigned int mp_dstapic;
>> -     unsigned char mp_type;
>> -     unsigned char mp_irqtype;
>> -     unsigned short mp_irqflag;
>> -     unsigned char mp_srcbus;
>> -     unsigned char mp_srcbusirq;
>> -     unsigned char mp_dstirq;
>> -};
>
> So you replaced that with mpc_ioapic and mpc_intsrc uses:
>
>  struct mpc_ioapic {
>         unsigned char type;
>         unsigned char apicid;
>         unsigned char apicver;
>         unsigned char flags;
>         unsigned int apicaddr;
>  };
>
>  struct mpc_intsrc {
>         unsigned char type;
>         unsigned char irqtype;
>         unsigned short irqflag;
>         unsigned char srcbus;
>         unsigned char srcbusirq;
>         unsigned char dstapic;
>         unsigned char dstirq;
>  };
>
> And removed the mp_config_ioapic and mp_config_intsrc types.
>
> The mp_config_ioapic and mp_config_intsrc types are kernel-internal, while
> mpc_ioapic and mpc_intsrc is an MP Specification type - the lowlevel
> structure of these tables as provided in RAM or ROM by the BIOS.
>
> Here's the potential problems that need to be thought through:
>
> - the biggest potential problem is that apicaddr is 32-bit wide while
>  mp_apicaddr is 64-bit wide, on the 64-bit kernel. Can an apic address be
>  above 4 GB, in an mptable? I dont think it can be.
>

mp_apicaddr is getting its value from apicaddr which is u32
(arch/x86/kernel/acpi/boot.c)
 OR
mp_apicaddr is getting its value from apicaddr which is u32
(arch/x86/kernel/mpparse.c)

> - another problem could be that dstapic are all 1-byte entities - while in
>  the in-kernel version (mp_apicid and mp_dstapic) they are 32 bit. The
>  ACPI code fills in the tables too so maybe there it can be wider than 8
>  bits?
>

mp_apicid is getting its value from uniq_ioapic_id(id) which returns
u8 (arch/x86/kernel/acpi/boot.c)
 OR
mp_apicid is getting its value from reg_00.bits.ID which also 8 bits
(arch/x86/kernel/io_apic.c)
 OR
mp_apicid is getting its value from apicid which also 8 bits
(arch/x86/kernel/mpparse.c)

and

mp_dstapic is getting its value from mp_apicid which is already 8 bits
(arch/x86/kernel/acpi/boot.c)
 OR
mp_dstapic is getting its value from mpc_intsrc->dstapic which is
already 8 bits (arch/x86/kernel/mpparse.c)

By using current structure we are wasting space as well as creating
confusion for checking u8 <-> u32 and u32 <-> u64.

Thanks,
--
JSR