Since we might be having COM32 and COM32R in the same filesystem, do
check for the appropriate COM32R magic before doing anything else.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

When entering 16-bit PM after shuffle and boot, we might as well
do so sanely.  Specifically, set up the data segments so that they
match the code segment, generating a 16-bit "tiny" model environment.

This makes it a lot saner to bootstrap a proper PM environment from
there if that is what the user intends.  For the presumably more
common case of RM entry, it won't do any harm, and it's only a handful
of additional instructions.

Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

Clarify that the CS.base restriction is a prescription of
compatibility, and not something that the ABI handles transparently.

Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

Add note about KVM restrictions on CS.base before entering real mode.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

KVM uses V86 mode to simulate real mode.  This causes problems with
the new shuffler.  This changes the shuffler handover to be in
16-bit protected mode instead, and requires the stub to do the actual
entry to real mode.  For the KVM hack to work, all segments must have:

(seg.base & 0xfff0000f) == 0 && seg.limit == 0xffff

As a result, we have to make sure the real-mode entry stub is
paragraph-aligned, lest we violate the first criterion.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

The new shuffler machinery is a major enough change that it's probably
a good idea to bump the version number to 3.80.

Conflicts:
	com32/lib/syslinux/memmap.c

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

Impact: should deal with memory holes more gracefully

Right now, if we find a memory hole, we simply don't use the memory
beyond that point.  This makes it possible for com32 modules to
use that memory.  After this, we should be able to add relocation
support to allow loading when the memory at 1 MB is unavailable.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

Make the new shuffler actually work.  This includes changing
rllpack.inc to run in 32-bit mode (since simple_pm_call now switches
to 32-bit mode) and changing the new shuffler interface to move the
shuffle list before actually doing any work.

Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>