{- ----------------------------------------------------------------------------- -- -- (c) The University of Glasgow 2001-2003 -- -- Access to system tools: gcc, cp, rm etc -- ----------------------------------------------------------------------------- -} {-# LANGUAGE CPP, MultiWayIf, ScopedTypeVariables #-} module SysTools ( -- * Initialisation initSysTools, lazyInitLlvmConfig, -- * Interface to system tools module SysTools.Tasks, module SysTools.Info, linkDynLib, copy, copyWithHeader, -- * General utilities Option(..), expandTopDir, -- * Platform-specifics libmLinkOpts, -- * Mac OS X frameworks getPkgFrameworkOpts, getFrameworkOpts ) where #include "HsVersions.h" import GhcPrelude import GHC.Settings import Module import Packages import Outputable import ErrUtils import GHC.Platform import DynFlags import Control.Monad.Trans.Except (runExceptT) import System.FilePath import System.IO import System.IO.Unsafe (unsafeInterleaveIO) import SysTools.ExtraObj import SysTools.Info import SysTools.Tasks import SysTools.BaseDir import SysTools.Settings {- Note [How GHC finds toolchain utilities] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SysTools.initSysProgs figures out exactly where all the auxiliary programs are, and initialises mutable variables to make it easy to call them. To do this, it makes use of definitions in Config.hs, which is a Haskell file containing variables whose value is figured out by the build system. Config.hs contains two sorts of things cGCC, The *names* of the programs cCPP e.g. cGCC = gcc cUNLIT cCPP = gcc -E etc They do *not* include paths cUNLIT_DIR The *path* to the directory containing unlit, split etc cSPLIT_DIR *relative* to the root of the build tree, for use when running *in-place* in a build tree (only) --------------------------------------------- NOTES for an ALTERNATIVE scheme (i.e *not* what is currently implemented): Another hair-brained scheme for simplifying the current tool location nightmare in GHC: Simon originally suggested using another configuration file along the lines of GCC's specs file - which is fine except that it means adding code to read yet another configuration file. What I didn't notice is that the current package.conf is general enough to do this: Package {name = "tools", import_dirs = [], source_dirs = [], library_dirs = [], hs_libraries = [], extra_libraries = [], include_dirs = [], c_includes = [], package_deps = [], extra_ghc_opts = ["-pgmc/usr/bin/gcc","-pgml${topdir}/bin/unlit", ... etc.], extra_cc_opts = [], extra_ld_opts = []} Which would have the advantage that we get to collect together in one place the path-specific package stuff with the path-specific tool stuff. End of NOTES --------------------------------------------- ************************************************************************ * * \subsection{Initialisation} * * ************************************************************************ -} -- Note [LLVM configuration] -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -- The `llvm-targets` and `llvm-passes` files are shipped with GHC and contain -- information needed by the LLVM backend to invoke `llc` and `opt`. -- Specifically: -- -- * llvm-targets maps autoconf host triples to the corresponding LLVM -- `data-layout` declarations. This information is extracted from clang using -- the script in utils/llvm-targets/gen-data-layout.sh and should be updated -- whenever we target a new version of LLVM. -- -- * llvm-passes maps GHC optimization levels to sets of LLVM optimization -- flags that GHC should pass to `opt`. -- -- This information is contained in files rather the GHC source to allow users -- to add new targets to GHC without having to recompile the compiler. -- -- Since this information is only needed by the LLVM backend we load it lazily -- with unsafeInterleaveIO. Consequently it is important that we lazily pattern -- match on LlvmConfig until we actually need its contents. lazyInitLlvmConfig :: String -> IO LlvmConfig lazyInitLlvmConfig top_dir = unsafeInterleaveIO $ do -- see Note [LLVM configuration] targets <- readAndParse "llvm-targets" mkLlvmTarget passes <- readAndParse "llvm-passes" id return $ LlvmConfig { llvmTargets = targets, llvmPasses = passes } where readAndParse name builder = do let llvmConfigFile = top_dir </> name llvmConfigStr <- readFile llvmConfigFile case maybeReadFuzzy llvmConfigStr of Just s -> return (fmap builder <$> s) Nothing -> pgmError ("Can't parse " ++ show llvmConfigFile) mkLlvmTarget :: (String, String, String) -> LlvmTarget mkLlvmTarget (dl, cpu, attrs) = LlvmTarget dl cpu (words attrs) initSysTools :: String -- TopDir path -> IO Settings -- Set all the mutable variables above, holding -- (a) the system programs -- (b) the package-config file -- (c) the GHC usage message initSysTools top_dir = do res <- runExceptT $ initSettings top_dir case res of Right a -> pure a Left (SettingsError_MissingData msg) -> pgmError msg Left (SettingsError_BadData msg) -> pgmError msg {- Note [Windows stack usage] See: #8870 (and #8834 for related info) and #12186 On Windows, occasionally we need to grow the stack. In order to do this, we would normally just bump the stack pointer - but there's a catch on Windows. If the stack pointer is bumped by more than a single page, then the pages between the initial pointer and the resulting location must be properly committed by the Windows virtual memory subsystem. This is only needed in the event we bump by more than one page (i.e 4097 bytes or more). Windows compilers solve this by emitting a call to a special function called _chkstk, which does this committing of the pages for you. The reason this was causing a segfault was because due to the fact the new code generator tends to generate larger functions, we needed more stack space in GHC itself. In the x86 codegen, we needed approximately ~12kb of stack space in one go, which caused the process to segfault, as the intervening pages were not committed. GCC can emit such a check for us automatically but only when the flag -fstack-check is used. See https://gcc.gnu.org/onlinedocs/gnat_ugn/Stack-Overflow-Checking.html for more information. -} copy :: DynFlags -> String -> FilePath -> FilePath -> IO () copy dflags purpose from to = copyWithHeader dflags purpose Nothing from to copyWithHeader :: DynFlags -> String -> Maybe String -> FilePath -> FilePath -> IO () copyWithHeader dflags purpose maybe_header from to = do showPass dflags purpose hout <- openBinaryFile to WriteMode hin <- openBinaryFile from ReadMode ls <- hGetContents hin -- inefficient, but it'll do for now. ToDo: speed up maybe (return ()) (header hout) maybe_header hPutStr hout ls hClose hout hClose hin where -- write the header string in UTF-8. The header is something like -- {-# LINE "foo.hs" #-} -- and we want to make sure a Unicode filename isn't mangled. header h str = do hSetEncoding h utf8 hPutStr h str hSetBinaryMode h True {- ************************************************************************ * * \subsection{Support code} * * ************************************************************************ -} linkDynLib :: DynFlags -> [String] -> [InstalledUnitId] -> IO () linkDynLib dflags0 o_files dep_packages = do let -- This is a rather ugly hack to fix dynamically linked -- GHC on Windows. If GHC is linked with -threaded, then -- it links against libHSrts_thr. But if base is linked -- against libHSrts, then both end up getting loaded, -- and things go wrong. We therefore link the libraries -- with the same RTS flags that we link GHC with. dflags1 = if platformMisc_ghcThreaded $ platformMisc dflags0 then addWay' WayThreaded dflags0 else dflags0 dflags2 = if platformMisc_ghcDebugged $ platformMisc dflags1 then addWay' WayDebug dflags1 else dflags1 dflags = updateWays dflags2 verbFlags = getVerbFlags dflags o_file = outputFile dflags pkgs <- getPreloadPackagesAnd dflags dep_packages let platform = targetPlatform dflags os = platformOS platform let pkg_lib_paths = collectLibraryPaths dflags pkgs let pkg_lib_path_opts = concatMap get_pkg_lib_path_opts pkg_lib_paths get_pkg_lib_path_opts l | ( osElfTarget (platformOS (targetPlatform dflags)) || osMachOTarget (platformOS (targetPlatform dflags)) ) && dynLibLoader dflags == SystemDependent && -- Only if we want dynamic libraries WayDyn `elem` ways dflags && -- Only use RPath if we explicitly asked for it useXLinkerRPath dflags os = ["-L" ++ l, "-Xlinker", "-rpath", "-Xlinker", l] -- See Note [-Xlinker -rpath vs -Wl,-rpath] | otherwise = ["-L" ++ l] let lib_paths = libraryPaths dflags let lib_path_opts = map ("-L"++) lib_paths -- We don't want to link our dynamic libs against the RTS package, -- because the RTS lib comes in several flavours and we want to be -- able to pick the flavour when a binary is linked. -- On Windows we need to link the RTS import lib as Windows does -- not allow undefined symbols. -- The RTS library path is still added to the library search path -- above in case the RTS is being explicitly linked in (see #3807). let pkgs_no_rts = case os of OSMinGW32 -> pkgs _ -> filter ((/= rtsUnitId) . packageConfigId) pkgs let pkg_link_opts = let (package_hs_libs, extra_libs, other_flags) = collectLinkOpts dflags pkgs_no_rts in package_hs_libs ++ extra_libs ++ other_flags -- probably _stub.o files -- and last temporary shared object file let extra_ld_inputs = ldInputs dflags -- frameworks pkg_framework_opts <- getPkgFrameworkOpts dflags platform (map unitId pkgs) let framework_opts = getFrameworkOpts dflags platform case os of OSMinGW32 -> do ------------------------------------------------------------- -- Making a DLL ------------------------------------------------------------- let output_fn = case o_file of Just s -> s Nothing -> "HSdll.dll" runLink dflags ( map Option verbFlags ++ [ Option "-o" , FileOption "" output_fn , Option "-shared" ] ++ [ FileOption "-Wl,--out-implib=" (output_fn ++ ".a") | gopt Opt_SharedImplib dflags ] ++ map (FileOption "") o_files -- Permit the linker to auto link _symbol to _imp_symbol -- This lets us link against DLLs without needing an "import library" ++ [Option "-Wl,--enable-auto-import"] ++ extra_ld_inputs ++ map Option ( lib_path_opts ++ pkg_lib_path_opts ++ pkg_link_opts )) _ | os == OSDarwin -> do ------------------------------------------------------------------- -- Making a darwin dylib ------------------------------------------------------------------- -- About the options used for Darwin: -- -dynamiclib -- Apple's way of saying -shared -- -undefined dynamic_lookup: -- Without these options, we'd have to specify the correct -- dependencies for each of the dylibs. Note that we could -- (and should) do without this for all libraries except -- the RTS; all we need to do is to pass the correct -- HSfoo_dyn.dylib files to the link command. -- This feature requires Mac OS X 10.3 or later; there is -- a similar feature, -flat_namespace -undefined suppress, -- which works on earlier versions, but it has other -- disadvantages. -- -single_module -- Build the dynamic library as a single "module", i.e. no -- dynamic binding nonsense when referring to symbols from -- within the library. The NCG assumes that this option is -- specified (on i386, at least). -- -install_name -- Mac OS/X stores the path where a dynamic library is (to -- be) installed in the library itself. It's called the -- "install name" of the library. Then any library or -- executable that links against it before it's installed -- will search for it in its ultimate install location. -- By default we set the install name to the absolute path -- at build time, but it can be overridden by the -- -dylib-install-name option passed to ghc. Cabal does -- this. ------------------------------------------------------------------- let output_fn = case o_file of { Just s -> s; Nothing -> "a.out"; } instName <- case dylibInstallName dflags of Just n -> return n Nothing -> return $ "@rpath" `combine` (takeFileName output_fn) runLink dflags ( map Option verbFlags ++ [ Option "-dynamiclib" , Option "-o" , FileOption "" output_fn ] ++ map Option o_files ++ [ Option "-undefined", Option "dynamic_lookup", Option "-single_module" ] ++ (if platformArch platform `elem` [ ArchX86_64, ArchAArch64 ] then [ ] else [ Option "-Wl,-read_only_relocs,suppress" ]) ++ [ Option "-install_name", Option instName ] ++ map Option lib_path_opts ++ extra_ld_inputs ++ map Option framework_opts ++ map Option pkg_lib_path_opts ++ map Option pkg_link_opts ++ map Option pkg_framework_opts -- dead_strip_dylibs, will remove unused dylibs, and thus save -- space in the load commands. The -headerpad is necessary so -- that we can inject more @rpath's later for the leftover -- libraries in the runInjectRpaths phase below. -- -- See Note [Dynamic linking on macOS] ++ [ Option "-Wl,-dead_strip_dylibs", Option "-Wl,-headerpad,8000" ] ) runInjectRPaths dflags pkg_lib_paths output_fn _ -> do ------------------------------------------------------------------- -- Making a DSO ------------------------------------------------------------------- let output_fn = case o_file of { Just s -> s; Nothing -> "a.out"; } unregisterised = platformUnregisterised (targetPlatform dflags) let bsymbolicFlag = -- we need symbolic linking to resolve -- non-PIC intra-package-relocations for -- performance (where symbolic linking works) -- See Note [-Bsymbolic assumptions by GHC] ["-Wl,-Bsymbolic" | not unregisterised] runLink dflags ( map Option verbFlags ++ libmLinkOpts ++ [ Option "-o" , FileOption "" output_fn ] ++ map Option o_files ++ [ Option "-shared" ] ++ map Option bsymbolicFlag -- Set the library soname. We use -h rather than -soname as -- Solaris 10 doesn't support the latter: ++ [ Option ("-Wl,-h," ++ takeFileName output_fn) ] ++ extra_ld_inputs ++ map Option lib_path_opts ++ map Option pkg_lib_path_opts ++ map Option pkg_link_opts ) -- | Some platforms require that we explicitly link against @libm@ if any -- math-y things are used (which we assume to include all programs). See #14022. libmLinkOpts :: [Option] libmLinkOpts = #if defined(HAVE_LIBM) [Option "-lm"] #else [] #endif getPkgFrameworkOpts :: DynFlags -> Platform -> [InstalledUnitId] -> IO [String] getPkgFrameworkOpts dflags platform dep_packages | platformUsesFrameworks platform = do pkg_framework_path_opts <- do pkg_framework_paths <- getPackageFrameworkPath dflags dep_packages return $ map ("-F" ++) pkg_framework_paths pkg_framework_opts <- do pkg_frameworks <- getPackageFrameworks dflags dep_packages return $ concat [ ["-framework", fw] | fw <- pkg_frameworks ] return (pkg_framework_path_opts ++ pkg_framework_opts) | otherwise = return [] getFrameworkOpts :: DynFlags -> Platform -> [String] getFrameworkOpts dflags platform | platformUsesFrameworks platform = framework_path_opts ++ framework_opts | otherwise = [] where framework_paths = frameworkPaths dflags framework_path_opts = map ("-F" ++) framework_paths frameworks = cmdlineFrameworks dflags -- reverse because they're added in reverse order from the cmd line: framework_opts = concat [ ["-framework", fw] | fw <- reverse frameworks ] {- Note [-Bsymbolic assumptions by GHC] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ GHC has a few assumptions about interaction of relocations in NCG and linker: 1. -Bsymbolic resolves internal references when the shared library is linked, which is important for performance. 2. When there is a reference to data in a shared library from the main program, the runtime linker relocates the data object into the main program using an R_*_COPY relocation. 3. If we used -Bsymbolic, then this results in multiple copies of the data object, because some references have already been resolved to point to the original instance. This is bad! We work around [3.] for native compiled code by avoiding the generation of R_*_COPY relocations. Unregisterised compiler can't evade R_*_COPY relocations easily thus we disable -Bsymbolic linking there. See related tickets: #4210, #15338 -}