Re: [PATCH] reserve end-of-conventional-memory to 1MB on 32-bit v2

[Jeremy Fitzhardinge <jeremy@goop.org>]

Alexander van Heukelum wrote:
> On Tue, 04 Mar 2008 07:18:48 -0800, "Jeremy Fitzhardinge"
> <jeremy@goop.org> said:
>   
>> Mark McLoughlin wrote:
>>     
>>> This is a bit magic, is it worth splitting it out as something like
>>> is_paravirt_environment() ?
>>>   
>>>       
>> Yes, is_paravirt() already exists for this purpose.
>>     
>
> Hi,
>
> If it exists, it is well-hidden. A grep for is_paravirt on the testing
> tree turns up nothing. Did you get the name right?
>   

Nope.  paravirt_enabled().

>> The code looking at the boot params will only work if we actually booted 
>> via the paravirt Linux boot protocol, which Xen doesn't at present.
>>     
>
> I chose to code it exactly this way, because it is what is used in
> head_32.S to choose how to start the kernel. Or is this code not
> executed at all?
>   

No, this isn't executed when booting under Xen.  Xen is a bit magic, and 
has its own kernel entrypoint which the domain builder finds via 
Xen-specific ELF notes on the vmlinux.  We plan to move to booting via 
this path at some point.

> [excerpt form head_32.S]
>         cmpw $0x207, pa(boot_params + BP_version)
>         jb default_entry
>
>         /* Paravirt-compatible boot parameters.  Look to see what
>         architecture
>                 we're booting under. */
>         movl pa(boot_params + BP_hardware_subarch), %eax
>         cmpl $num_subarch_entries, %eax
>         jae bad_subarch
>
>         movl pa(subarch_entries)(,%eax,4), %eax
>         subl $__PAGE_OFFSET, %eax
>         jmp *%eax
> [and]
> subarch_entries:
>         .long default_entry             /* normal x86/PC */
>         .long lguest_entry              /* lguest hypervisor */
>         .long xen_entry                 /* Xen hypervisor */
> num_subarch_entries = (. - subarch_entries) / 4
> [end]
>
> If this is indeed not executed, is there a way to detect whether
> we can expect the environment to behave like a normal pc in terms
> of magic addresses, bios areas, isa reserved address space and so
> on?
>   

Just because something is paravirtualized and uses a non-PC 
hardware_subarch doesn't mean this stuff isn't present.  Even the 
paravirt_enabled() test isn't accurate, because the environment may 
still choose to emulate these things (or in the Xen dom0 case, it may 
directly expose the real hardware).

    J