Re: [PATCH] x86/HVM: rewrite the start info structure definition in binary form

["Roger Pau Monné" <roger.pau@citrix.com>]

El 9/2/16 a les 9:14, Jan Beulich ha escrit:
>>>> On 05.02.16 at 16:45, <roger.pau@citrix.com> wrote:
>> El 5/2/16 a les 14:13, Jan Beulich ha escrit:
>>> But
>>> even then - wouldn't what I've read on the other thread mean
>>> at least the filename should be put there (as kind of the first
>>> command line element)?
>>
>> Really? I didn't get that impression at all, what's the filename useful
>> for anyway?
> 
> This largely depends on whether you mean to mimic multiboot1 or
> multiboot2. Remember the placeholder people have to add to
> certain grub2 invocation lines? I think that's a result of their
> attempt to had through a file name. But indeed - I've never
> understood what it's good for until this request came about: How
> else would the consumer know which module is which if there's no
> signature or alike inside the file? Iirc someone suggested (in the
> context of the discussion here) that Linux might look for "initrd"
> in the filename to identify that one. Seems fragile to me, but may
> be the only alternative if not demanding multiple modules to be
> ordered in a certain way.

OK, right now xl/libxl only allows passing one module that's considered
the initram/ramdisk from Linux PoV. Other OSes that use the pv loader
don't pass any module at all AFAIK. The problem is that there's no way
to specify a command line parameter for modules, and I think in order to
make use of this we should add such option, instead of hard coding
"initrd" as the command line of the first module.

However, I don't think any of this should prevent this patch from being
accepted. If someone has interest in implementing a way to pass module
command line arguments please feel free to do it, this patch just sets
the right ABI for doing so, although it's unused at the moment.

Roger.