linker script: implement SIZEOF_HEADERS

[emaste]

Bugzilla Link	28688
Resolution	FIXED
Resolved on	Aug 10, 2016 11:43
Version	unspecified
OS	FreeBSD
Blocks	#23588
CC	@rui314

Extended Description

SIZEOF_HEADERS is used by the linker script in FreeBSD's linux emulation layer:

SECTIONS
{
. = . + SIZEOF_HEADERS;

    .hash           : { *(.hash) }                  :text
    .gnu.hash       : { *(.gnu.hash) }

...

SIZEOF_HEADERS' sizeof_headers'
Return the size in bytes of the output file's headers. You can
use this number as the start address of the first section, if you
choose, to facilitate paging.

[llvmbot]

Take.

[emaste]

Also used in the main FreeBSD kernel link script.
https://svnweb.freebsd.org/base/head/sys/conf/ldscript.amd64?view=markup#l10

[llvmbot]

So, I'm taking a shot at implementing this but I'd like to discuss the implementation a little bit before digging into code.
SIZEOF_HEADERS should be set equal to ehdr_size + phdr_size * segment count.
The issue here is that we don't know exactly the number of segments that we're going to create because the SECTION clause of the linker script might change that.
My idea is to have a preprocessing step, parsing the linker script in order to get the section lists from it (and use that to compute the correct value of SIZEOF_HEADERS).
This code changed quite a bit recently, os I might miss something here. Comments appreciated =)

[rui314]

SIZEOF_HEADERS seems pretty tricky to me and doesn't seem to be recommended to use without PHDRS command. I think it's a reasonable stance.

https://sourceware.org/binutils/docs/ld/Builtin-Functions.html#Builtin-Functions

I'm wondering what the FreeBSD's linker script is doing with SIZEOF_HEADERS. I don't understand why it is needed for Xen. Could you give me an insight?

[emaste]

I'm wondering what the FreeBSD's linker script is doing with SIZEOF_HEADERS. I > don't understand why it is needed for Xen. Could you give me an insight?

It's not for just for Xen; SIZEOF_HEADERS has been there since the beginning of the FreeBSD/amd64 project. For reference the default (built-in) userland linker script in GNU ld uses it like so:

SECTIONS
{
/* Read-only sections, merged into text segment: */
PROVIDE (__executable_start = 0x400000); . = 0x400000 + SIZEOF_HEADERS;

and the kernel use is equivalent:

SECTIONS
{
/* Read-only sections, merged into text segment: */
kernphys = CONSTANT (MAXPAGESIZE);
. = kernbase + kernphys + SIZEOF_HEADERS;

Xen support added the "AT (kernphys + SIZEOF_HEADERS)" but that use is no different really.

I don't really have an objection to the binutils restrictions:

To avoid this error, you must avoid using the SIZEOF_HEADERS function, or you must rework your linker script to avoid forcing the linker to use additional program headers, or you must define the program headers yourself using the PHDRS command (see PHDRS).

except that I think "additional program headers" in the GNU ld case is based on the default layout it assumes, which we don't really want to do.

For the FreeBSD kernel we could certainly add PHDRS if that's the best way forward.

[rui314]

I'm inclined to allow SIZEOF_HEADER only when PHDRS commands are used to give us a full definition of the program headers. It is simple and easy to understand. Otherwise, we could either (1) assume some PHDRS and report an error if the program needs more PHDRS or (2) walk over all the SECTIONS command twice as Davide said. I think both are a bit too much for this variable.

[llvmbot]

I'm inclined to allow SIZEOF_HEADER only when PHDRS commands are used to
give us a full definition of the program headers. It is simple and easy to
understand. Otherwise, we could either (1) assume some PHDRS and report an
error if the program needs more PHDRS or (2) walk over all the SECTIONS
command twice as Davide said. I think both are a bit too much for this
variable.

At fact we are already walk over SECTIONS twice. First pass is createSections(), the second is assignAddresses(). Can't we use that ? I would try to estimate segments amount during the first pass.
And we need SIZEOF_HEADER only on second pass when we assign addresses I think.

[emaste]

Review: https://reviews.llvm.org/D23165
Commit: https://reviews.llvm.org/rL278204