Re: [PATCH 1 of 2] Add the support for Xen to include OVMF UEFI support and directly use it

[Attilio Rao <attilio.rao@citrix.com>]

On 23/02/12 10:07, Ian Campbell wrote:
> On Wed, 2012-02-22 at 19:17 +0000, Attilio Rao wrote:
>    
>> A way to integrate OVMF build directly into XEN has still be discussed
>> on the mailing list appropriately.
>>      
> AIUI OVMF is maintained in SVN. Our normal procedure for adding an
> external dependency would be for us to mirror it on xenbits as a
> convenience to our users, who don't need to get stuff from multiple
> places, and as a courtesy to our upstreams, so our users don't consume
> their resources.
>
> I don't much fancy setting the necessary webdav or whatever stuff on
> xenbits and integrating SVN support into our build system though. What
> do people think about using git-svn to manage our mirror in git instead?
> Or better: perhaps OVMF have an official git or hg mirror?
>
> Anyone have any thoughts/opinions/better ideas etc?
>
>    
>> diff -r a88ba599add1 -r 032fea10f8d1 tools/firmware/hvmloader/config.h
>> --- a/tools/firmware/hvmloader/config.h	Tue Feb 21 17:45:59 2012 +0000
>> +++ b/tools/firmware/hvmloader/config.h	Wed Feb 22 18:54:03 2012 +0000
>> @@ -35,6 +35,8 @@ struct bios_config {
>>
>>   extern struct bios_config rombios_config;
>>   extern struct bios_config seabios_config;
>> +extern struct bios_config ovmf32_config;
>> +extern struct bios_config ovmf64_config;
>>      
> Can you confirm that you need an OVMF which matches the OS bit-width you
> are installing. i..e that there is no support for booting a 32 bit EFI
> OS (or bootloader, shell, whatever it is called) on a 64 bit OVMF?
>
>    

I didn't test this case, really, but I would think OVMF-64 / OS-32 could 
possibly work.
You are suggesting if this is the case we should just ship the 64-bit 
emulation?
> [...]
>    
>> +static void ovmf_acpi_build_tables(void)
>> +{
>> +    struct acpi_config config = {
>> +        .dsdt_anycpu = dsdt_anycpu,
>> +        .dsdt_anycpu_len = dsdt_anycpu_len,
>> +        .dsdt_15cpu = dsdt_15cpu,
>> +        .dsdt_15cpu_len = dsdt_15cpu_len,
>> +    };
>>      
> IIRC the 15cpu tables are there to workaround a bug in some old version
> of Windows (2k?). I think therefore you can omit these on the basis that
> no version of Windows with that bug also supports EFI. seabios.c does
> this too.
>    

Yes, I was just unsure which of the 2 approaches (use dsdt_15cpu/skip 
completely) was preferred here so I went with the more 'complete' but it 
makes sense.

Thanks,
Attilio