2MB of zeros in kernel file

[Taylor R Campbell <campbell+netbsd-port-amd64%mumble.net@localhost>]

The NetBSD/amd64 kernel file has 2 MB of zeros near the beginning.
Why?

In October 2009, nick@ added `-z max-page-size=0x10000' to
Makefile.amd64 (r1.26/1.27), with the following comment:

> MAXPAGESIZE got bumped to 0x200000. Use -z maxpagesize=0x100000 until
> someone(tm) decides that kernels should have that alignment, etc.

Later, in February 2017, maxv@ bumped it to 0x20000 (r1.57):

> Put 2MB alignments between the kernel segments. This way the kernel image
> is entirely mapped with large pages, which uniformizes performance and
> reduces fluctuation. Sent on port-amd64.

The message `sent on port-amd64' was presumably:
https://mail-index.NetBSD.org/port-amd64/2017/02/03/msg002496.html
(although the link there no longer works so I cannot 100% confirm that
the patch proposed is what was committed)

These changes have the effect of putting 2 MB of zeros at the
beginning of the netbsd kernel file, as you can confirm with hexdump:

% hexdump -C netbsd
00000000  7f 45 4c 46 02 01 01 00  00 00 00 00 00 00 00 00  |.ELF............|
00000010  02 00 3e 00 01 00 00 00  00 90 20 80 ff ff ff ff  |..>....... .....|
00000020  40 00 00 00 00 00 00 00  a8 cb ad 01 00 00 00 00  |@...............|
00000030  00 00 00 00 40 00 38 00  02 00 40 00 1d 00 1c 00  |....@.8...@.....|
00000040  01 00 00 00 05 00 00 00  00 00 20 00 00 00 00 00  |.......... .....|
00000050  00 00 20 80 ff ff ff ff  00 00 20 00 00 00 00 00  |.. ....... .....|
00000060  28 16 42 01 00 00 00 00  28 16 42 01 00 00 00 00  |(.B.....(.B.....|
00000070  00 00 20 00 00 00 00 00  01 00 00 00 06 00 00 00  |.. .............|
00000080  00 00 80 01 00 00 00 00  00 00 80 81 ff ff ff ff  |................|
00000090  00 00 80 01 00 00 00 00  e0 7a 0a 00 00 00 00 00  |.........z......|
000000a0  00 00 20 00 00 00 00 00  00 00 20 00 00 00 00 00  |.. ....... .....|
000000b0  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00200000  0f 01 f8 48 a3 08 00 00  00 00 ff ff ff 48 a1 10  |...H.........H..|
...

I am not sure offhand whether the 2 MB of zeros get loaded into
memory, but I've been sitting on a patch from a year ago to replace it
by max-page-size=0x1000 with the effect of shaving those 2 MB of zeros
off the file.

I don't remember investigating the consequences of setting
max-page-size=0x1000 beyond shaving the 2 MB of zeros off the file;
it's possible I got stuck trying to get the large-page advantage
without the 2 MB of zeros by tweaking the ldscript instead.

Perhaps this isn't actually an issue if we always deal in compressed
kernel files, because gzip will handily compress a long run of zeros.
But I thought I'd open a discussion here so I can stop sitting on this
patch.