Re: [PATCH 2/7] multiboot2: Allow 64-bit entry tags

[Roger Pau Monné]

On Thu, Mar 28, 2024 at 03:05:47PM +0000, Ross Lagerwall wrote:
> On Tue, Mar 19, 2024 at 10:07 AM Roger Pau Monné <roger.pau@citrix.com> wrote:
> >
> > On Wed, Mar 13, 2024 at 03:07:43PM +0000, Ross Lagerwall wrote:
> > > Binaries may be built with entry points above 4G. While bootloaders may
> > > relocate them below 4G, it should be possible for the binary to specify
> > > those entry points. Therefore, extend the multiboot2 protocol such that
> > > 64 bit addresses are allowed for entry points. The extension is done in
> > > a backwards-compatible way.
> > >
> > > Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> > > ---
> > >  doc/multiboot.texi | 32 +++++++++++++++++++-------------
> > >  doc/multiboot2.h   |  6 +++++-
> > >  2 files changed, 24 insertions(+), 14 deletions(-)
> > >
> > > diff --git a/doc/multiboot.texi b/doc/multiboot.texi
> > > index d12719c744eb..049afab53c1f 100644
> > > --- a/doc/multiboot.texi
> > > +++ b/doc/multiboot.texi
> > > @@ -522,12 +522,12 @@ header.
> > >
> > >  @example
> > >  @group
> > > -        +-------------------+
> > > -u16     | type = 3          |
> > > -u16     | flags             |
> > > -u32     | size              |
> > > -u32     | entry_addr        |
> > > -        +-------------------+
> > > +          +-------------------+
> > > +u16       | type = 3          |
> > > +u16       | flags             |
> > > +u32       | size              |
> > > +u32 / u64 | entry_addr        |
> > > +          +-------------------+
> >
> > I might be confused, but this entry point is used in 32bit protected
> > mode, and hence a 64bit value is simply impossible to use according to
> > the protocol in "3.3 I386 machine state".
> >
> > Unless that section is expanded to describe other protocols that use
> > the entry address in a way where 64bits could be meaningful it seems
> > pointless to expand the field.
> 
> I changed this because the same binary is being used for both BIOS boot
> and UEFI boot, therefore it may have a base address above 4 GiB.
> Despite that, it is expected that GRUB would relocate the binary below
> 4 GiB so BIOS boot would still work.

Right, for UEFI boot it's possible to have entry addresses above 4GB,
because the entry point is called in long mode with identity page
tables (and hence you can put addresses in %rip past the 4GB
boundary).

However the multiboot entry point puts the CPU in 32bit protected
mode, and hence %eip can only hold a value below the 4GB boundary.

It's technically impossible to use an entry point above 4GB, unless
there's something that I'm missing that changes the initial CPU state
for the multiboot2 entry point.

Thanks, Roger.