I'm trying to see if it's possible to use a gcc plugin in an ARM cross compiler (arm-none-eabi-gcc). I'm running into compiler errors however, and am questioning whether what I'm trying to do is possible.
The plugin I'm trying to set up is: https://github.com/vanhauser-thc/AFLplusplus/tree/master/gcc_plugin
I'm compiling the plugin on x86-64 linux using the -m32 flag, since the cross compiler is a 32-bit application. However when I try to use the plugin in the cross compiler using -fplugin , I get an undefined symbol compiler error:
cc1plus: error: cannot load plugin ../afl-gcc-pass.so
../afl-gcc-pass.so: undefined symbol: _Z13build_int_cstP9tree_nodel
I looked through the plugin's symbols using nm and discovered that the majority of symbols are undefined, including ones like exit and random. I'm new to most of this and am unsure of what that really means. Some searching online suggested that it may have had something to do with incorrect library paths, but setting LIBRARY_PATH and LD_LIBRARY_PATH and rebuilding did not seem to help.
The gcc version set-ups I have tried:
1: x86: 5.4.0 , arm: 5.4.1 on ubuntu 16.04
2: x86: 5.2.0 , arm: 5.2.1 on CentOS 6.8
Is it possible to use a gcc plugin in a different gcc than it was compiled with or am I wasting my time?
Yes, it is possible to build a gcc plugin with a given compiler and then use the plugin in another compiler (including a cross-compiler), but you have to make sure you include the right header files when building the plugin.
Specifically, you have to include the plugin development header files of the target compiler, instead of those of the host compiler. The directory where plugin development files for your target compiler are located can be obtained with the following command:
$(TARGET_CC) -print-file-name=plugin
where $(TARGET_CC) is your target compiler. So a concise way to specify the relevant include directory in the compiler flags when building the plugin would be something like -I"$(shell $(TARGET_CC) -print-file-name=plugin)/include".
For the specific plugin you are trying to use (instrumentation for afl-fuzz), in order to build the plugin for your cross-compiler you could modify the Makefile in the gcc_plugin folder; more specifically, you could define a TARGET_CC variable containing the path to your cross-compiler, and then replace $(CC) with $(TARGET_CC) in the definition of PLUGIN_FLAGS, as in:
PLUGIN_FLAGS = -fPIC -fno-rtti -I"$(shell $(TARGET_CC) -print-file-name=plugin)/include"
You will also have to comment out the commands executed in the test_build Makefile target, because those commands would try to use the plugin with the native compiler and so would fail.
Then, you will be able to use the plugin with your cross-compiler, as in:
arm-none-eabi-gcc -fplugin=../afl-gcc-pass.so --specs=nosys.specs my_source_file.c
Related
I have a cc_library (tbb) that requires the compiler flag -mwaitpkg on some compilers (Clang) to compile successfully. At the same time, there are older versions of GCC (4.9) that do not know this flag, and therefore the compilation via GCC 4.9 leads to an error:
gcc: error: unrecognized command line option '-mwaitpkg'
In a more advanced Bazel setup, I guess one would work around this using hermetic toolchains. This way every toolchain could provide its own set of compiler flags. Nevertheless, I do not want to enforce any specific toolchain and I am not sure if this is the right way to go (move copts to toolchain?).
Also introducing a config would be a way to solve this problem. E.g. bazel build --config=waitpkg //.... But this would require that a user is aware of this config and also knows the details of using waitpkg.
What is a proper "Bazel-way" to handle different compiler flags for different compilers?
The flag '-mwaitpkg' is supported by GCC version 9.3, Clang* 12, and newer versions of those tools.
If you build Bazel with earlier versions of GCC, you should remove the flag otherwise it gives compilation errors.
I'm trying to compile CMake using a non-default GCC installed in /usr/local/gcc530, on Solaris 2.11.
I have LD_LIBRARY_PATH=/usr/local/gcc530/lib/sparcv9
Bootstrap proceeds fine, bootstrapped cmake successfully compiles various object files, but when it tries to link the real cmake (and other executables), I get pages of "undefined reference" errors to various standard library functions, because, as running the link command manually with -Wl,-verbose shows, the linker links with /usr/lib/64/libstdc++.so of the system default, much older GCC.
This is because apparently CMake tries to find curses/ncurses libraries (even if I tell it BUILD_CursesDialog:BOOL=OFF), finds them in /usr/lib/64, and adds -L/usr/lib/64 to build/Source/CMakeFiles/cmake.dir/link.txt, which causes the linker to use libstdc++.so from there, and not my actual GCC's own.
I found a workaround: I can get the path to proper libraries from $CC -m64 -print-file-name=libstdc++.so then put it with -L into LDFLAGS when running ./configure, and all works well then.
Is there a less hacky way? It's really weird that I can't tell GCC to prioritize its own libraries.
Also, is there some way to have CMake explain where different parts of a resulting command line came from?
I am trying to bootstrap GCC during OpenELEC compilation. I need to add libatomic for the target system so as to compile some packages.
When I try to add libatomic for the target system, I get a compilation error with:
/home/mathieu/tmp/OpenELEC.tv/build.OpenELEC-ci20.mips-8.0-devel/toolchain/mipsel-openelec-linux-gnu/bin/ld: cannot find crti.o: No such file or directory
/home/mathieu/tmp/OpenELEC.tv/build.OpenELEC-ci20.mips-8.0-devel/toolchain/mipsel-openelec-linux-gnu/bin/ld: cannot find -lc
/home/mathieu/tmp/OpenELEC.tv/build.OpenELEC-ci20.mips-8.0-devel/toolchain/mipsel-openelec-linux-gnu/bin/ld: cannot find crtn.o: No such file or directory
collect2: error: ld returned 1 exit status
Using the following package.mk file:
https://github.com/malaterre/OpenELEC.tv/blob/fd5a5558104ed38aee1c53bb6d31ba73e8eb6e57/packages/lang/gcc/package.mk
If that matter I am targetting a MIPS system, specifically the Creator CI20:
https://github.com/malaterre/OpenELEC.tv/blob/master/config/arch.mips
I am not clear about the OpenELEC build system, but it seems as if the host compiler and target compiler are build at the same time, while I would need to build binutils in between:
http://preshing.com/20141119/how-to-build-a-gcc-cross-compiler/
You might need to have a GCC built without libatomic to build one with libatomic. The intermediary compiler will serve to build your CRT (if you do not have it already) and libatomic, so that a second compiler can use these "prebuilt" things.
That was not hard at all. Basically one need to create first a minimal gcc (bootstrap) and then build the full one with libatomic:
https://github.com/malaterre/OpenELEC.tv/blob/70dbe25c1647f01eb83c108939c470437a2db259/packages/lang/gcc/package.mk
I have some fortran programs that would not compile in old versions of gfortran. I have to run multiple instances of this program and am using another system (a cluster system) which has centos5_x64 with gcc-4.1 !!
Therefore I had to build new version of gcc; I built both gcc-4.8.3 and gcc-4.9.2 in my home folder. These programs use hdf5 and so the latter also has to be compiled using the same compiler. I tested the fortran programs after removing the hdf5 dependency on both gfortran-4.8.3 and 4.8.9 and they get built and execute properly. I also tested simple C/C++ programs (with basic i/o and arithmetic) with the new gcc/g++(s); they work fine. Before compiling hdf5 libraries I set these environment variables:
PATH=<GCCPATH>:$PATH
LD_LIBRARY_PATH=<GCCLIB>
LD_RUN_PATH=<GCCLIB>
export PATH
export LD_LIBRARY_PATH
export LD_RUN_PATH
HDF5 specific instructions
4.3.7. Specifying other libraries and headers
Configure searches the standard places (those places known by the
systems compiler) for include files and header files. However,
additional directories can be specified by using the CPPFLAGS
and/or LDFLAGS variables:
$ CPPFLAGS=-I/home/robb/include \
LDFLAGS=-L/home/robb/lib \
LDFLAGS=-L<GCCLIB>
CPPFLAGS=-I<GCCINCLUDE>
then configured as:
./configure --prefix=$HOME/HDF5 --enable-fortran
During make I get this error:
/usr/lib/../lib64/crt1.o: In function `_start':
(.text+0x20): undefined reference to `main'
collect2: error: ld returned 1 exit status
This happens with both versions of gcc but not with the versions installed in the standard location (This happened in another system too which had gcc 4.4 installed). With general searching I got to know that this error is associated with the absence of a main() and in such cases -c flag has to be passed. However all examples of this error were from people's personal scripts and were not for libraries. Please let me know if I am missing something.
Upgrading the system OS is not a choice as of now.
As indicated in the comment by doqtor, I modified the Makefile post configure. It seems that CPPFLAGS and LDFLAGS were not set.
After setting the right CPPFLAGS and LDFLAGS in the Makefile, the compilation happened successfully.
So after searching the web for a while, Ive decided to try here as it seems to be a good forum for discussion. Im trying to create a simple gcc plugin. The program code is attached in the end of this mail, but in plain english it registers the plugin and makes sure that the pragma_init function is called when pragmas are registered. It is here that I use c_register_pragma to intercept some of the pragmas.
I compile it using the example in http://gcc.gnu.org/onlinedocs/gccint/Plugins-building.html#Plugins-building. The compilation and linking works fine. However, when I load the plug-in I get:
gcc -c -fplugin=plugin.so test.c -o test.o
cc1: error: cannot load plugin plugin.so
plugin.so: undefined symbol: warning
What am I doing wrong? In addition, when including some header files (that will be required later), I get a lot of errors. For example, including "tree.h" yields (amongst 50 other errors):
/machmode.h:262:1: error: unknown type name 'class'
class bit_field_mode_iterator
^
/machmode.h:263:1: error: expected '=', ',', ';', 'asm' or '__attribute__' before '{' token
{
^
/plugin/include/tree.h:27:0,
from conftest.c:63:
/vec.h:220:8: error: field 'register_overhead' declared as a function
Anyone have a clue on what I am doing wrong?
Thank you
There are two problems here :
The error : "cannot load plugin plugin.so" means that you should add to your LD_LIBRARY_PATH the directory where you store your new shared library plugin.
The hundreds of errors you got with all the files in the include are resolved in my computer if you compile with g++ instead of gcc (not sure to understand why thought)
Which version of GCC are you using, both to compile your plugin, and to use the plugin? Run simply
gcc -v
without any other program argument to find out!
Did you install the appropriate package for GCC plugin development (on Debian or Ubuntu, it might be gcc-4.7-plugin-dev, but adapt the 4.7 version to your particular version of GCC)?
Did you install all the dependencies needed to build your GCC (on Debian or Ubuntu, apt-get build-dep gcc-4.7 gcc-4.7-plugin-dev)?
Recent versions of GCC (notably many GCC 4.7 shipped by distributions, and all GCC 4.8) are compiled by a C++ compiler, not a C compiler.
You may check how was your GCC built (in C or in C++) by running
nm -D -C $(gcc -print-file-name=cc1)
If that command shows typed C++ manged names, e.g. execute_ipa_pass_list(opt_pass*) instead of just execute_ipa_pass_list your GCC has been compiled with a C++ compiler (probably g++)
So you may need to use g++ (not gcc) to compile your GCC plugin.
As I commented, did you consider using MELT (a domain specific language to extend GCC) to extend or customize your gcc compiler?
I suggest downloading the very latest http://gcc-melt.org/melt-plugin-snapshot.tar.bz2 since I will release the next MELT in a few weeks for GCC 4.7 and 4.8
And don't expect to change the parsing behavior of your GCC with a plugin. That is not really possible (GCC provides only plugin hooks to add your builtins and pragmas, not to extend the parsed syntax).