Moving early_recursion_flag (4 bytes) after early_level4_pgt (4k) and
early_dynamic_pgts (256k) saves 4k which are used for alignment of
early_level4_pgt after early_recursion_flag.

The real improvement is merely on the source code side. Previously it
was:
* __INITDATA + .balign
* early_recursion_flag variable
* a ton of CPP MACROS
* __INITDATA (again)
* early_top_pgt and early_recursion_flag variables
* .data

Now, it is a bit simpler:
* a ton of CPP MACROS
* __INITDATA + .balign
* early_top_pgt and early_recursion_flag variables
* early_recursion_flag variable
* .data

On the binary level the change looks like this:
Before:
 (sections)
  12 .init.data    00042000  0000000000000000  0000000000000000 00008000  2**12
 (symbols)
  000000       4 OBJECT  GLOBAL DEFAULT   22 early_recursion_flag
  001000    4096 OBJECT  GLOBAL DEFAULT   22 early_top_pgt
  002000 0x40000 OBJECT  GLOBAL DEFAULT   22 early_dynamic_pgts

After:
 (sections)
  12 .init.data    00041004  0000000000000000  0000000000000000 00008000  2**12
 (symbols)
  000000    4096 OBJECT  GLOBAL DEFAULT   22 early_top_pgt
  001000 0x40000 OBJECT  GLOBAL DEFAULT   22 early_dynamic_pgts
  041000       4 OBJECT  GLOBAL DEFAULT   22 early_recursion_flag

So the resulting vmlinux is smaller by 4k with my toolchain as many
other variables can be placed after early_recursion_flag to fill the
rest of the page. Note that this is only .init data, so it is freed
right after being booted anyway. Savings on-disk are none -- compression
of zeros is easy, so the size of bzImage is the same pre and post the
change.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86-ml <x86@kernel.org>
Link: https://lkml.kernel.org/r/20191003095238.29831-1-jslaby@suse.cz