GNU-centric view of toolchains

Beginner's introduction to GCC

About me

maya@NetBSD.org

coypu@sdf.org

NetBSD/pkgsrc for the last 3 years

(Not a GCC expert, just sharing knowledge)

This talk

GNU-centric view of toolchains

(alternatives exist, won't be mentioned)

Might be familiar for many

let's all get on the same page

Top-level overview of toolchain

preprocessor

compiler

assembler

linker

Top-level overview of toolchain

preprocessor

compiler

assembler

linker

rtld (ld.so)

Upstreams overview

project	component
GCC	preprocessor
GCC	compiler
binutils	assembler
binutils	linker
OS	rtld (ld.so)

Independent tools

command	component
cpp	preprocessor
/usr/libexec/cc1	compiler
as	assembler
ld	linker
(kernel)	rtld (ld.so)

GCC can pass flags to each component

GCC flag	component
-Wp,	preprocessor
(none)	compiler
-Wa,	assembler
-Wl,	linker

Can stop after component

GCC flag	stop at
-E	preprocessor
-S	compiler
-c	assembler
(none)	linker

Preprocessor

important for packaging, most of our problems are here

Expands preprocessor directives

#include <math.h>
#if defined(__NetBSD__) || defined(__linux__)...
#error "OS not in long list of supported OSes"

#include <math.h> ?

(why would things "just work"?)

Existing lookup directories
Visible with -Wp,--verbose
Each tool has --verbose

#include <math.h>

~> gcc -Wp,--verbose test.c
#include "..." search starts here:
#include <...> search starts here:
 /usr/include/gcc-6
 /usr/include
End of search list.

I have /usr/include/math.h

#if defined(__NetBSD__) ?

I never defined that...

GCC internal code builtin_define("__NetBSD__")

visible with gcc -dM -E - < /dev/null

~> gcc -dM -E - < /dev/null
#define __NetBSD__ 1
#define _LP64 1
#define __STDC_VERSION__ 201112L
...

Results can be inspected after preprocessing

-save-temps to save all the results

Single file with all the used code
Standalone test case for compiler problems
No need to dig through many include files
quickly debug your qt5+boost program with limited prior knowledge

(the reason for this talk)

compiler

?? Magical C to assembler machine ??

assembler

Human-like assembly output
Some directives

	.file	"test2.c"
	.text
	.globl	main
	.type	main, @function
main:
.LFB0:
	.cfi_startproc
	pushq	%rbp
	.cfi_def_cfa_offset 16
	.cfi_offset 6, -16
	movq	%rsp, %rbp
	.cfi_def_cfa_register 6
	movl	$0, %eax
	popq	%rbp
	.cfi_def_cfa 7, 8
	ret
	.cfi_endproc
.LFE0:
	.size	main, .-main
	.ident	"GCC: (nb2 20180327) 6.4.0"

Linker/object files

Machine-readable
Many tools to parse

Parsing binary objects (nm)

~> nm test2
0000000000600a78 d _DYNAMIC
0000000000600bf8 d _GLOBAL_OFFSET_TABLE_
                 w _Jv_RegisterClasses
0000000000600a58 D __CTOR_LIST_END__
0000000000400968 r __GNU_EH_FRAME_HDR
00000000004006ea T ___start
0000000000600c92 B __bss_start
                 w __deregister_frame_info
0000000000600c48 D __dso_handle
0000000000600c40 D __progname
0000000000600c98 B __ps_strings
                 w __register_frame_info
00000000004005e8 r __rela_iplt_end
00000000004005e8 r __rela_iplt_start
0000000000400690 T __start
                 U __syscall
0000000000600c92 D _edata
0000000000600cb0 B _end
                 U _exit
00000000004008d0 T _fini
00000000004005f0 T _init
                 U _libc_init
0000000000400690 T _start
                 U abort
                 U atexit
0000000000600ca8 B environ
                 U exit
00000000004008c1 T main

C runtime stuff

(where all these extra symbols came from?)

We can see that the linker adding those

C runtime stuff

~> gcc -Wl,--verbose test2.c
.. (linker script) ..
attempt to open /usr/lib/crt0.o succeeded
/usr/lib/crt0.o
attempt to open /usr/lib/crti.o succeeded
/usr/lib/crti.o
attempt to open /usr/lib/crtbegin.o succeeded
/usr/lib/crtbegin.o
attempt to open /var/tmp//ccrus4oG.o succeeded
/var/tmp//ccrus4oG.o
attempt to open /usr/lib/libgcc_s.so succeeded
-lgcc_s (/usr/lib/libgcc_s.so)
attempt to open /usr/lib/libgcc.so failed
attempt to open /usr/lib/libgcc.a succeeded
attempt to open /usr/lib/libc.so succeeded
-lc (/usr/lib/libc.so)
attempt to open /usr/lib/libgcc_s.so succeeded
-lgcc_s (/usr/lib/libgcc_s.so)
attempt to open /usr/lib/libgcc.so failed
attempt to open /usr/lib/libgcc.a succeeded
attempt to open /usr/lib/crtend.o succeeded
/usr/lib/crtend.o
attempt to open /usr/lib/crtn.o succeeded
/usr/lib/crtn.o

specfiles

Not very legible, gcc -dumpspecs

Responsible for not needing to specify -lc

attempt to open /usr/lib/libc.so succeeded
-lc (/usr/lib/libc.so)
attempt to open /usr/lib/libgcc_s.so succeeded
-lgcc_s (/usr/lib/libgcc_s.so)

specfiles

*lib:
%{pthread:			     %{!p:			       %{!pg:-lpthread}}	     %{p:-lpthread_p}		     %{pg:-lpthread_p}}		   %{posix:			     %{!p:			       %{!pg:-lposix}}		     %{p:-lposix_p}		     %{pg:-lposix_p}}		   %{shared:-lc}		   %{!shared:			     %{!symbolic:		       %{!p:				 %{!pg:-lc}}		       %{p:-lc_p}		       %{pg:-lc_p}}}

Parsing binary objects (readelf)

Can see the libraries we use

Dynamic section at offset 0x2adc0 contains 22 entries:
  Tag        Type                         Name/Value
 0x0000000000000001 (NEEDED)             Shared library: [libedit.so.3]
 0x0000000000000001 (NEEDED)             Shared library: [libterminfo.so.1]
 0x0000000000000001 (NEEDED)             Shared library: [libc.so.12]
 0x000000000000000f (RPATH)              Library rpath: [/lib]

Semver

We can see we specify libc.MAJOR

libc.MAJOR -> libc.MAJOR.minor (symlink)

Can change library minor without binary noticing
New library major won't be used

GCC configuration

Primary file: gcc/config.gcc
per-OS+arch
Add headers to override definitions
Add extra makefiles for extra code to run e.g. on init
(e.g. NetBSD has custom code to rename cabs to __c99_cabs

GCC - overview

front-end	GIMPLE
middle-end	RTL
back-end	assembly

Can dump lots of intermediate results

-fdump-rtl-all-all
-fdump-tree-all

Backend (RTL)

gcc/config/ARCH/arch.md
lisp-like, rules to match

Backend (RTL)

constraints must match
multiple templates possible

(define_insn "extendqihi2"
  [(set (match_operand:HI 0 "nonimmediate_operand" "=g")
        (sign_extend:HI (match_operand:QI 1 "nonimmediate_operand" "g")))]
  ""
  "cvtbw %1,%0")

Backend (RTL)

Want to reorder assembly
Need to specify side-effects (condition codes)
(x86 FLAGS)

(define_insn_and_split "*cmp_cc_i387"
  [(set (reg:CCFP FLAGS_REG)
        (compare:CCFP
          (match_operand:MODEF 1 "register_operand" "f")
          (match_operand:MODEF 2 "nonimmediate_operand" "fm")))
   (clobber (match_operand:HI 0 "register_operand" "=a"))]
  "TARGET_80387 && TARGET_SAHF && !TARGET_CMOVE"
  "#"
  "&& reload_completed"
  [(set (match_dup 0)
        (unspec:HI
          [(compare:CCFP (match_dup 1)(match_dup 2))]
        UNSPEC_FNSTSW))
   (set (reg:CC FLAGS_REG)
        (unspec:CC [(match_dup 0)] UNSPEC_SAHF))]
  ""
  [(set_attr "type" "multi")
   (set_attr "unit" "i387")
   (set_attr "mode" "")])