Linker.c

/* -----------------------------------------------------------------------------
 * $Id: Linker.c,v 1.137 2003/10/08 10:37:25 wolfgang Exp $
 *
 * (c) The GHC Team, 2000-2003
 *
 * RTS Object Linker
 *
 * ---------------------------------------------------------------------------*/

#if 0
#include "PosixSource.h"
#endif
#include "Rts.h"
#include "RtsFlags.h"
#include "HsFFI.h"
#include "Hash.h"
#include "Linker.h"
#include "LinkerInternals.h"
#include "RtsUtils.h"
#include "StoragePriv.h"
#include "Schedule.h"

#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif

#include <stdlib.h>
#include <string.h>

#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif

#if defined(HAVE_FRAMEWORK_HASKELLSUPPORT)
#include <HaskellSupport/dlfcn.h>
#elif defined(HAVE_DLFCN_H)
#include <dlfcn.h>
#endif

#if defined(cygwin32_TARGET_OS)
#ifdef HAVE_DIRENT_H
#include <dirent.h>
#endif

#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#include <regex.h>
#include <sys/fcntl.h>
#include <sys/termios.h>
#include <sys/utime.h>
#include <sys/utsname.h>
#include <sys/wait.h>
#endif

#if defined(ia64_TARGET_ARCH)
#define USE_MMAP
#include <fcntl.h>
#include <sys/mman.h>
#endif

#if defined(linux_TARGET_OS) || defined(solaris2_TARGET_OS) || defined(freebsd_TARGET_OS) || defined(netbsd_TARGET_OS) || defined(openbsd_TARGET_OS)
#  define OBJFORMAT_ELF
#elif defined(cygwin32_TARGET_OS) || defined (mingw32_TARGET_OS)
#  define OBJFORMAT_PEi386
#  include <windows.h>
#  include <math.h>
#elif defined(darwin_TARGET_OS)
#  include <mach-o/ppc/reloc.h>
#  define OBJFORMAT_MACHO
#  include <mach-o/loader.h>
#  include <mach-o/nlist.h>
#  include <mach-o/reloc.h>
#  include <mach-o/dyld.h>
#endif

/* Hash table mapping symbol names to Symbol */
static /*Str*/HashTable *symhash;

/* List of currently loaded objects */
ObjectCode *objects = NULL;	/* initially empty */

#if defined(OBJFORMAT_ELF)
static int ocVerifyImage_ELF    ( ObjectCode* oc );
static int ocGetNames_ELF       ( ObjectCode* oc );
static int ocResolve_ELF        ( ObjectCode* oc );
#elif defined(OBJFORMAT_PEi386)
static int ocVerifyImage_PEi386 ( ObjectCode* oc );
static int ocGetNames_PEi386    ( ObjectCode* oc );
static int ocResolve_PEi386     ( ObjectCode* oc );
#elif defined(OBJFORMAT_MACHO)
static int ocAllocateJumpIslands_MachO ( ObjectCode* oc );
static int ocVerifyImage_MachO    ( ObjectCode* oc );
static int ocGetNames_MachO       ( ObjectCode* oc );
static int ocResolve_MachO        ( ObjectCode* oc );

static void machoInitSymbolsWithoutUnderscore( void );
#endif

/* -----------------------------------------------------------------------------
 * Built-in symbols from the RTS
 */

typedef struct _RtsSymbolVal {
    char   *lbl;
    void   *addr;
} RtsSymbolVal;


#if !defined(PAR)
#define Maybe_ForeignObj        SymX(mkForeignObjzh_fast)

#define Maybe_Stable_Names      SymX(mkWeakzh_fast)			\
      				SymX(makeStableNamezh_fast)		\
      				SymX(finalizzeWeakzh_fast)
#else
/* These are not available in GUM!!! -- HWL */
#define Maybe_ForeignObj
#define Maybe_Stable_Names
#endif

#if !defined (mingw32_TARGET_OS)
#define RTS_POSIX_ONLY_SYMBOLS                  \
      SymX(stg_sig_install)			\
      Sym(nocldstop)
#endif

#if defined (cygwin32_TARGET_OS)
#define RTS_MINGW_ONLY_SYMBOLS /**/
/* Don't have the ability to read import libs / archives, so
 * we have to stupidly list a lot of what libcygwin.a
 * exports; sigh. 
 */
#define RTS_CYGWIN_ONLY_SYMBOLS                 \
      SymX(regfree)                             \
      SymX(regexec)                             \
      SymX(regerror)                            \
      SymX(regcomp)                             \
      SymX(__errno)                             \
      SymX(access)                              \
      SymX(chmod)                               \
      SymX(chdir)                               \
      SymX(close)                               \
      SymX(creat)                               \
      SymX(dup)                                 \
      SymX(dup2)                                \
      SymX(fstat)                               \
      SymX(fcntl)                               \
      SymX(getcwd)                              \
      SymX(getenv)                              \
      SymX(lseek)                               \
      SymX(open)                                \
      SymX(fpathconf)                           \
      SymX(pathconf)                            \
      SymX(stat)                                \
      SymX(pow)                                 \
      SymX(tanh)                                \
      SymX(cosh)                                \
      SymX(sinh)                                \
      SymX(atan)                                \
      SymX(acos)                                \
      SymX(asin)                                \
      SymX(tan)                                 \
      SymX(cos)                                 \
      SymX(sin)                                 \
      SymX(exp)                                 \
      SymX(log)                                 \
      SymX(sqrt)                                \
      SymX(localtime_r)                         \
      SymX(gmtime_r)                            \
      SymX(mktime)                              \
      Sym(_imp___tzname)                        \
      SymX(gettimeofday)                        \
      SymX(timezone)                            \
      SymX(tcgetattr)                           \
      SymX(tcsetattr)                           \
      SymX(memcpy)                              \
      SymX(memmove)                             \
      SymX(realloc)                             \
      SymX(malloc)                              \
      SymX(free)                                \
      SymX(fork)                                \
      SymX(lstat)                               \
      SymX(isatty)                              \
      SymX(mkdir)                               \
      SymX(opendir)                             \
      SymX(readdir)                             \
      SymX(rewinddir)                           \
      SymX(closedir)                            \
      SymX(link)                                \
      SymX(mkfifo)                              \
      SymX(pipe)                                \
      SymX(read)                                \
      SymX(rename)                              \
      SymX(rmdir)                               \
      SymX(select)                              \
      SymX(system)                              \
      SymX(write)                               \
      SymX(strcmp)                              \
      SymX(strcpy)                              \
      SymX(strncpy)                             \
      SymX(strerror)                            \
      SymX(sigaddset)                           \
      SymX(sigemptyset)                         \
      SymX(sigprocmask)                         \
      SymX(umask)                               \
      SymX(uname)                               \
      SymX(unlink)                              \
      SymX(utime)                               \
      SymX(waitpid)

#elif !defined(mingw32_TARGET_OS)
#define RTS_MINGW_ONLY_SYMBOLS /**/
#define RTS_CYGWIN_ONLY_SYMBOLS /**/
#else /* defined(mingw32_TARGET_OS) */
#define RTS_POSIX_ONLY_SYMBOLS  /**/
#define RTS_CYGWIN_ONLY_SYMBOLS /**/

/* Extra syms gen'ed by mingw-2's gcc-3.2: */
#if __GNUC__>=3
#define RTS_MINGW_EXTRA_SYMS                    \
      Sym(_imp____mb_cur_max)                   \
      Sym(_imp___pctype)            
#else
#define RTS_MINGW_EXTRA_SYMS
#endif

/* These are statically linked from the mingw libraries into the ghc
   executable, so we have to employ this hack. */
#define RTS_MINGW_ONLY_SYMBOLS                  \
      SymX(asyncReadzh_fast)			\
      SymX(asyncWritezh_fast)			\
      SymX(memset)                              \
      SymX(inet_ntoa)                           \
      SymX(inet_addr)                           \
      SymX(htonl)                               \
      SymX(recvfrom)                            \
      SymX(listen)                              \
      SymX(bind)                                \
      SymX(shutdown)                            \
      SymX(connect)                             \
      SymX(htons)                               \
      SymX(ntohs)                               \
      SymX(getservbyname)                       \
      SymX(getservbyport)                       \
      SymX(getprotobynumber)                    \
      SymX(getprotobyname)                      \
      SymX(gethostbyname)                       \
      SymX(gethostbyaddr)                       \
      SymX(gethostname)                         \
      SymX(strcpy)                              \
      SymX(strncpy)                             \
      SymX(abort)                               \
      Sym(_alloca)                              \
      Sym(isxdigit)                             \
      Sym(isupper)                              \
      Sym(ispunct)                              \
      Sym(islower)                              \
      Sym(isspace)                              \
      Sym(isprint)                              \
      Sym(isdigit)                              \
      Sym(iscntrl)                              \
      Sym(isalpha)                              \
      Sym(isalnum)                              \
      SymX(strcmp)                              \
      SymX(memmove)                             \
      SymX(realloc)                             \
      SymX(malloc)                              \
      SymX(pow)                                 \
      SymX(tanh)                                \
      SymX(cosh)                                \
      SymX(sinh)                                \
      SymX(atan)                                \
      SymX(acos)                                \
      SymX(asin)                                \
      SymX(tan)                                 \
      SymX(cos)                                 \
      SymX(sin)                                 \
      SymX(exp)                                 \
      SymX(log)                                 \
      SymX(sqrt)                                \
      SymX(memcpy)                              \
      Sym(mktime)                               \
      Sym(_imp___timezone)                      \
      Sym(_imp___tzname)                        \
      Sym(_imp___iob)                           \
      Sym(localtime)                            \
      Sym(gmtime)                               \
      Sym(opendir)                              \
      Sym(readdir)                              \
      Sym(rewinddir)                            \
      RTS_MINGW_EXTRA_SYMS                      \
      Sym(closedir)
#endif

#ifndef SMP
# define MAIN_CAP_SYM SymX(MainCapability)
#else
# define MAIN_CAP_SYM
#endif

#define RTS_SYMBOLS				\
      Maybe_ForeignObj				\
      Maybe_Stable_Names			\
      Sym(StgReturn)				\
      SymX(stg_enter_info)			\
      SymX(stg_enter_ret)			\
      SymX(stg_gc_void_info)			\
      SymX(__stg_gc_enter_1)			\
      SymX(stg_gc_noregs)			\
      SymX(stg_gc_unpt_r1_info)			\
      SymX(stg_gc_unpt_r1)			\
      SymX(stg_gc_unbx_r1_info)			\
      SymX(stg_gc_unbx_r1)			\
      SymX(stg_gc_f1_info)			\
      SymX(stg_gc_f1)				\
      SymX(stg_gc_d1_info)			\
      SymX(stg_gc_d1)				\
      SymX(stg_gc_l1_info)			\
      SymX(stg_gc_l1)				\
      SymX(__stg_gc_fun)			\
      SymX(stg_gc_fun_info)			\
      SymX(stg_gc_fun_ret)			\
      SymX(stg_gc_gen)				\
      SymX(stg_gc_gen_info)			\
      SymX(stg_gc_gen_hp)			\
      SymX(stg_gc_ut)				\
      SymX(stg_gen_yield)			\
      SymX(stg_yield_noregs)			\
      SymX(stg_yield_to_interpreter)		\
      SymX(stg_gen_block)			\
      SymX(stg_block_noregs)			\
      SymX(stg_block_1)				\
      SymX(stg_block_takemvar)			\
      SymX(stg_block_putmvar)			\
      SymX(stg_seq_frame_info)			\
      SymX(ErrorHdrHook)			\
      MAIN_CAP_SYM                              \
      SymX(MallocFailHook)			\
      SymX(OnExitHook)				\
      SymX(OutOfHeapHook)			\
      SymX(PatErrorHdrHook)			\
      SymX(PostTraceHook)			\
      SymX(PreTraceHook)			\
      SymX(StackOverflowHook)			\
      SymX(__encodeDouble)			\
      SymX(__encodeFloat)			\
      SymX(__gmpn_gcd_1)			\
      SymX(__gmpz_cmp)				\
      SymX(__gmpz_cmp_si)			\
      SymX(__gmpz_cmp_ui)			\
      SymX(__gmpz_get_si)			\
      SymX(__gmpz_get_ui)			\
      SymX(__int_encodeDouble)			\
      SymX(__int_encodeFloat)			\
      SymX(andIntegerzh_fast)			\
      SymX(blockAsyncExceptionszh_fast)		\
      SymX(catchzh_fast)			\
      SymX(cmp_thread)				\
      SymX(complementIntegerzh_fast)		\
      SymX(cmpIntegerzh_fast)	        	\
      SymX(cmpIntegerIntzh_fast)	      	\
      SymX(createAdjustor)			\
      SymX(decodeDoublezh_fast)			\
      SymX(decodeFloatzh_fast)			\
      SymX(defaultsHook)			\
      SymX(delayzh_fast)			\
      SymX(deRefWeakzh_fast)			\
      SymX(deRefStablePtrzh_fast)		\
      SymX(divExactIntegerzh_fast)		\
      SymX(divModIntegerzh_fast)		\
      SymX(forkzh_fast)				\
      SymX(forkProcess)				\
      SymX(forkOS_createThread)			\
      SymX(freeHaskellFunctionPtr)		\
      SymX(freeStablePtr)		        \
      SymX(gcdIntegerzh_fast)			\
      SymX(gcdIntegerIntzh_fast)		\
      SymX(gcdIntzh_fast)			\
      SymX(genSymZh)				\
      SymX(getProgArgv)				\
      SymX(getStablePtr)			\
      SymX(int2Integerzh_fast)			\
      SymX(integer2Intzh_fast)			\
      SymX(integer2Wordzh_fast)			\
      SymX(isCurrentThreadBoundzh_fast)		\
      SymX(isDoubleDenormalized)		\
      SymX(isDoubleInfinite)			\
      SymX(isDoubleNaN)				\
      SymX(isDoubleNegativeZero)		\
      SymX(isEmptyMVarzh_fast)			\
      SymX(isFloatDenormalized)			\
      SymX(isFloatInfinite)			\
      SymX(isFloatNaN)				\
      SymX(isFloatNegativeZero)			\
      SymX(killThreadzh_fast)			\
      SymX(makeStablePtrzh_fast)		\
      SymX(minusIntegerzh_fast)			\
      SymX(mkApUpd0zh_fast)			\
      SymX(myThreadIdzh_fast)			\
      SymX(labelThreadzh_fast)                  \
      SymX(newArrayzh_fast)			\
      SymX(newBCOzh_fast)			\
      SymX(newByteArrayzh_fast)			\
      SymX_redirect(newCAF, newDynCAF)		\
      SymX(newMVarzh_fast)			\
      SymX(newMutVarzh_fast)			\
      SymX(atomicModifyMutVarzh_fast)		\
      SymX(newPinnedByteArrayzh_fast)		\
      SymX(orIntegerzh_fast)			\
      SymX(performGC)				\
      SymX(plusIntegerzh_fast)			\
      SymX(prog_argc)				\
      SymX(prog_argv)				\
      SymX(putMVarzh_fast)			\
      SymX(quotIntegerzh_fast)			\
      SymX(quotRemIntegerzh_fast)		\
      SymX(raisezh_fast)			\
      SymX(raiseIOzh_fast)			\
      SymX(remIntegerzh_fast)			\
      SymX(resetNonBlockingFd)			\
      SymX(resumeThread)			\
      SymX(rts_apply)				\
      SymX(rts_checkSchedStatus)		\
      SymX(rts_eval)				\
      SymX(rts_evalIO)				\
      SymX(rts_evalLazyIO)			\
      SymX(rts_evalStableIO)			\
      SymX(rts_eval_)				\
      SymX(rts_getBool)				\
      SymX(rts_getChar)				\
      SymX(rts_getDouble)			\
      SymX(rts_getFloat)			\
      SymX(rts_getInt)				\
      SymX(rts_getInt32)			\
      SymX(rts_getPtr)				\
      SymX(rts_getFunPtr)			\
      SymX(rts_getStablePtr)			\
      SymX(rts_getThreadId)			\
      SymX(rts_getWord)				\
      SymX(rts_getWord32)			\
      SymX(rts_lock)				\
      SymX(rts_mkBool)				\
      SymX(rts_mkChar)				\
      SymX(rts_mkDouble)			\
      SymX(rts_mkFloat)				\
      SymX(rts_mkInt)				\
      SymX(rts_mkInt16)				\
      SymX(rts_mkInt32)				\
      SymX(rts_mkInt64)				\
      SymX(rts_mkInt8)				\
      SymX(rts_mkPtr)				\
      SymX(rts_mkFunPtr)			\
      SymX(rts_mkStablePtr)			\
      SymX(rts_mkString)			\
      SymX(rts_mkWord)				\
      SymX(rts_mkWord16)			\
      SymX(rts_mkWord32)			\
      SymX(rts_mkWord64)			\
      SymX(rts_mkWord8)				\
      SymX(rts_unlock)				\
      SymX(rtsSupportsBoundThreads)		\
      SymX(run_queue_hd)			\
      SymX(__hscore_get_saved_termios)		\
      SymX(__hscore_set_saved_termios)		\
      SymX(setProgArgv)				\
      SymX(startupHaskell)			\
      SymX(shutdownHaskell)			\
      SymX(shutdownHaskellAndExit)		\
      SymX(stable_ptr_table)			\
      SymX(stackOverflow)			\
      SymX(stg_CAF_BLACKHOLE_info)		\
      SymX(stg_BLACKHOLE_BQ_info)		\
      SymX(awakenBlockedQueue)			\
      SymX(stg_CHARLIKE_closure)		\
      SymX(stg_EMPTY_MVAR_info)			\
      SymX(stg_IND_STATIC_info)			\
      SymX(stg_INTLIKE_closure)			\
      SymX(stg_MUT_ARR_PTRS_FROZEN_info)	\
      SymX(stg_WEAK_info)                       \
      SymX(stg_ap_v_info)			\
      SymX(stg_ap_f_info)			\
      SymX(stg_ap_d_info)			\
      SymX(stg_ap_l_info)			\
      SymX(stg_ap_n_info)			\
      SymX(stg_ap_p_info)			\
      SymX(stg_ap_pv_info)			\
      SymX(stg_ap_pp_info)			\
      SymX(stg_ap_ppv_info)			\
      SymX(stg_ap_ppp_info)			\
      SymX(stg_ap_pppp_info)			\
      SymX(stg_ap_ppppp_info)			\
      SymX(stg_ap_pppppp_info)			\
      SymX(stg_ap_ppppppp_info)			\
      SymX(stg_ap_0_ret)			\
      SymX(stg_ap_v_ret)			\
      SymX(stg_ap_f_ret)			\
      SymX(stg_ap_d_ret)			\
      SymX(stg_ap_l_ret)			\
      SymX(stg_ap_n_ret)			\
      SymX(stg_ap_p_ret)			\
      SymX(stg_ap_pv_ret)			\
      SymX(stg_ap_pp_ret)			\
      SymX(stg_ap_ppv_ret)			\
      SymX(stg_ap_ppp_ret)			\
      SymX(stg_ap_pppp_ret)			\
      SymX(stg_ap_ppppp_ret)			\
      SymX(stg_ap_pppppp_ret)			\
      SymX(stg_ap_ppppppp_ret)			\
      SymX(stg_ap_1_upd_info)			\
      SymX(stg_ap_2_upd_info)			\
      SymX(stg_ap_3_upd_info)			\
      SymX(stg_ap_4_upd_info)			\
      SymX(stg_ap_5_upd_info)			\
      SymX(stg_ap_6_upd_info)			\
      SymX(stg_ap_7_upd_info)			\
      SymX(stg_ap_8_upd_info)			\
      SymX(stg_exit)				\
      SymX(stg_sel_0_upd_info)			\
      SymX(stg_sel_10_upd_info)			\
      SymX(stg_sel_11_upd_info)			\
      SymX(stg_sel_12_upd_info)			\
      SymX(stg_sel_13_upd_info)			\
      SymX(stg_sel_14_upd_info)			\
      SymX(stg_sel_15_upd_info)			\
      SymX(stg_sel_1_upd_info)			\
      SymX(stg_sel_2_upd_info)			\
      SymX(stg_sel_3_upd_info)			\
      SymX(stg_sel_4_upd_info)			\
      SymX(stg_sel_5_upd_info)			\
      SymX(stg_sel_6_upd_info)			\
      SymX(stg_sel_7_upd_info)			\
      SymX(stg_sel_8_upd_info)			\
      SymX(stg_sel_9_upd_info)			\
      SymX(stg_upd_frame_info)			\
      SymX(suspendThread)			\
      SymX(takeMVarzh_fast)			\
      SymX(timesIntegerzh_fast)			\
      SymX(tryPutMVarzh_fast)			\
      SymX(tryTakeMVarzh_fast)			\
      SymX(unblockAsyncExceptionszh_fast)	\
      SymX(unsafeThawArrayzh_fast)		\
      SymX(waitReadzh_fast)			\
      SymX(waitWritezh_fast)			\
      SymX(word2Integerzh_fast)			\
      SymX(xorIntegerzh_fast)			\
      SymX(yieldzh_fast)

#ifdef SUPPORT_LONG_LONGS
#define RTS_LONG_LONG_SYMS			\
      SymX(int64ToIntegerzh_fast)		\
      SymX(word64ToIntegerzh_fast)
#else
#define RTS_LONG_LONG_SYMS /* nothing */
#endif

// 64-bit support functions in libgcc.a
#if defined(__GNUC__) && SIZEOF_VOID_P <= 4
#define RTS_LIBGCC_SYMBOLS			\
      Sym(__divdi3)                             \
      Sym(__udivdi3)                            \
      Sym(__moddi3)                             \
      Sym(__umoddi3)				\
      Sym(__ashldi3)				\
      Sym(__ashrdi3)				\
      Sym(__lshrdi3)				\
      Sym(__eprintf)
#elif defined(ia64_TARGET_ARCH)
#define RTS_LIBGCC_SYMBOLS			\
      Sym(__divdi3)				\
      Sym(__udivdi3)                            \
      Sym(__moddi3)				\
      Sym(__umoddi3)				\
      Sym(__divsf3)				\
      Sym(__divdf3)
#else
#define RTS_LIBGCC_SYMBOLS
#endif

#ifdef darwin_TARGET_OS
      // Symbols that don't have a leading underscore
      // on Mac OS X. They have to receive special treatment,
      // see machoInitSymbolsWithoutUnderscore()
#define RTS_MACHO_NOUNDERLINE_SYMBOLS		\
      Sym(saveFP)				\
      Sym(restFP)
#endif

/* entirely bogus claims about types of these symbols */
#define Sym(vvv)  extern void vvv(void);
#define SymX(vvv) /**/
#define SymX_redirect(vvv,xxx) /**/
RTS_SYMBOLS
RTS_LONG_LONG_SYMS
RTS_POSIX_ONLY_SYMBOLS
RTS_MINGW_ONLY_SYMBOLS
RTS_CYGWIN_ONLY_SYMBOLS
RTS_LIBGCC_SYMBOLS
#undef Sym
#undef SymX
#undef SymX_redirect

#ifdef LEADING_UNDERSCORE
#define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
#else
#define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
#endif

#define Sym(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
                    (void*)(&(vvv)) },
#define SymX(vvv) Sym(vvv)

// SymX_redirect allows us to redirect references to one symbol to
// another symbol.  See newCAF/newDynCAF for an example.
#define SymX_redirect(vvv,xxx) \
    { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
      (void*)(&(xxx)) },

static RtsSymbolVal rtsSyms[] = {
      RTS_SYMBOLS
      RTS_LONG_LONG_SYMS
      RTS_POSIX_ONLY_SYMBOLS
      RTS_MINGW_ONLY_SYMBOLS
      RTS_CYGWIN_ONLY_SYMBOLS
      RTS_LIBGCC_SYMBOLS
      { 0, 0 } /* sentinel */
};

/* -----------------------------------------------------------------------------
 * Insert symbols into hash tables, checking for duplicates.
 */
static void ghciInsertStrHashTable ( char* obj_name,
                                     HashTable *table,
                                     char* key,
                                     void *data
				   )
{
   if (lookupHashTable(table, (StgWord)key) == NULL)
   {
      insertStrHashTable(table, (StgWord)key, data);
      return;
   }
   fprintf(stderr,
      "\n\n"
      "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
      "   %s\n"
      "whilst processing object file\n"
      "   %s\n"
      "This could be caused by:\n"
      "   * Loading two different object files which export the same symbol\n"
      "   * Specifying the same object file twice on the GHCi command line\n"
      "   * An incorrect `package.conf' entry, causing some object to be\n"
      "     loaded twice.\n"
      "GHCi cannot safely continue in this situation.  Exiting now.  Sorry.\n"
      "\n",
      (char*)key,
      obj_name
   );
   exit(1);
}


/* -----------------------------------------------------------------------------
 * initialize the object linker
 */


static int linker_init_done = 0 ;

#if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
static void *dl_prog_handle;
#endif

void
initLinker( void )
{
    RtsSymbolVal *sym;

    /* Make initLinker idempotent, so we can call it
       before evey relevant operation; that means we
       don't need to initialise the linker separately */
    if (linker_init_done == 1) { return; } else {
      linker_init_done = 1;
    }

    symhash = allocStrHashTable();

    /* populate the symbol table with stuff from the RTS */
    for (sym = rtsSyms; sym->lbl != NULL; sym++) {
	ghciInsertStrHashTable("(GHCi built-in symbols)",
                               symhash, sym->lbl, sym->addr);
    }
#   if defined(OBJFORMAT_MACHO)
    machoInitSymbolsWithoutUnderscore();
#   endif

#   if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
    dl_prog_handle = dlopen(NULL, RTLD_LAZY);
#   endif
}

/* -----------------------------------------------------------------------------
 *                  Loading DLL or .so dynamic libraries
 * -----------------------------------------------------------------------------
 *
 * Add a DLL from which symbols may be found.  In the ELF case, just
 * do RTLD_GLOBAL-style add, so no further messing around needs to
 * happen in order that symbols in the loaded .so are findable --
 * lookupSymbol() will subsequently see them by dlsym on the program's
 * dl-handle.  Returns NULL if success, otherwise ptr to an err msg.
 *
 * In the PEi386 case, open the DLLs and put handles to them in a
 * linked list.  When looking for a symbol, try all handles in the
 * list.  This means that we need to load even DLLs that are guaranteed
 * to be in the ghc.exe image already, just so we can get a handle
 * to give to loadSymbol, so that we can find the symbols.  For such
 * libraries, the LoadLibrary call should be a no-op except for returning
 * the handle.
 * 
 */

#if defined(OBJFORMAT_PEi386)
/* A record for storing handles into DLLs. */

typedef
   struct _OpenedDLL {
      char*              name;
      struct _OpenedDLL* next;
      HINSTANCE instance;
   }
   OpenedDLL;

/* A list thereof. */
static OpenedDLL* opened_dlls = NULL;
#endif

char *
addDLL( char *dll_name )
{
#  if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
   /* ------------------- ELF DLL loader ------------------- */
   void *hdl;
   char *errmsg;

   initLinker();

#if !defined(openbsd_TARGET_OS)
   hdl= dlopen(dll_name, RTLD_NOW | RTLD_GLOBAL);
#else
   hdl= dlopen(dll_name, RTLD_LAZY);
#endif
   if (hdl == NULL) {
      /* dlopen failed; return a ptr to the error msg. */
      errmsg = dlerror();
      if (errmsg == NULL) errmsg = "addDLL: unknown error";
      return errmsg;
   } else {
      return NULL;
   }
   /*NOTREACHED*/

#  elif defined(OBJFORMAT_PEi386)
   /* ------------------- Win32 DLL loader ------------------- */

   char*      buf;
   OpenedDLL* o_dll;
   HINSTANCE  instance;

   initLinker();

   /* fprintf(stderr, "\naddDLL; dll_name = `%s'\n", dll_name); */

   /* See if we've already got it, and ignore if so. */
   for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
      if (0 == strcmp(o_dll->name, dll_name))
         return NULL;
   }

   /* The file name has no suffix (yet) so that we can try
      both foo.dll and foo.drv

      The documentation for LoadLibrary says:
      	If no file name extension is specified in the lpFileName
      	parameter, the default library extension .dll is
      	appended. However, the file name string can include a trailing
      	point character (.) to indicate that the module name has no
      	extension. */

   buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
   sprintf(buf, "%s.DLL", dll_name);
   instance = LoadLibrary(buf);
   if (instance == NULL) {
	 sprintf(buf, "%s.DRV", dll_name);	// KAA: allow loading of drivers (like winspool.drv)
	 instance = LoadLibrary(buf);
	 if (instance == NULL) {
		stgFree(buf);

	    /* LoadLibrary failed; return a ptr to the error msg. */
	    return "addDLL: unknown error";
   	 }
   }
   stgFree(buf);

   /* Add this DLL to the list of DLLs in which to search for symbols. */
   o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
   o_dll->name     = stgMallocBytes(1+strlen(dll_name), "addDLL");
   strcpy(o_dll->name, dll_name);
   o_dll->instance = instance;
   o_dll->next     = opened_dlls;
   opened_dlls     = o_dll;

   return NULL;
#  else
   barf("addDLL: not implemented on this platform");
#  endif
}

/* -----------------------------------------------------------------------------
 * lookup a symbol in the hash table
 */
void *
lookupSymbol( char *lbl )
{
    void *val;
    initLinker() ;
    ASSERT(symhash != NULL);
    val = lookupStrHashTable(symhash, lbl);

    if (val == NULL) {
#       if defined(OBJFORMAT_ELF)
	return dlsym(dl_prog_handle, lbl);
#       elif defined(OBJFORMAT_MACHO)
	if(NSIsSymbolNameDefined(lbl)) {
	    NSSymbol symbol = NSLookupAndBindSymbol(lbl);
	    return NSAddressOfSymbol(symbol);
	} else {
	    return NULL;
	}
#       elif defined(OBJFORMAT_PEi386)
        OpenedDLL* o_dll;
        void* sym;
        for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
	  /* fprintf(stderr, "look in %s for %s\n", o_dll->name, lbl); */
           if (lbl[0] == '_') {
              /* HACK: if the name has an initial underscore, try stripping
                 it off & look that up first. I've yet to verify whether there's
                 a Rule that governs whether an initial '_' *should always* be
                 stripped off when mapping from import lib name to the DLL name.
              */
              sym = GetProcAddress(o_dll->instance, (lbl+1));
              if (sym != NULL) {
		/*fprintf(stderr, "found %s in %s\n", lbl+1,o_dll->name); fflush(stderr);*/
		return sym;
	      }
           }
           sym = GetProcAddress(o_dll->instance, lbl);
           if (sym != NULL) {
	     /*fprintf(stderr, "found %s in %s\n", lbl,o_dll->name); fflush(stderr);*/
	     return sym;
	   }
        }
        return NULL;
#       else
        ASSERT(2+2 == 5);
        return NULL;
#       endif
    } else {
	return val;
    }
}

static
__attribute((unused))
void *
lookupLocalSymbol( ObjectCode* oc, char *lbl )
{
    void *val;
    initLinker() ;
    val = lookupStrHashTable(oc->lochash, lbl);

    if (val == NULL) {
        return NULL;
    } else {
	return val;
    }
}


/* -----------------------------------------------------------------------------
 * Debugging aid: look in GHCi's object symbol tables for symbols
 * within DELTA bytes of the specified address, and show their names.
 */
#ifdef DEBUG
void ghci_enquire ( char* addr );

void ghci_enquire ( char* addr )
{
   int   i;
   char* sym;
   char* a;
   const int DELTA = 64;
   ObjectCode* oc;

   initLinker();

   for (oc = objects; oc; oc = oc->next) {
      for (i = 0; i < oc->n_symbols; i++) {
         sym = oc->symbols[i];
         if (sym == NULL) continue;
         // fprintf(stderr, "enquire %p %p\n", sym, oc->lochash);
         a = NULL;
         if (oc->lochash != NULL) {
            a = lookupStrHashTable(oc->lochash, sym);
	 }
         if (a == NULL) {
            a = lookupStrHashTable(symhash, sym);
	 }
         if (a == NULL) {
	     // fprintf(stderr, "ghci_enquire: can't find %s\n", sym);
         }
         else if (addr-DELTA <= a && a <= addr+DELTA) {
            fprintf(stderr, "%p + %3d  ==  `%s'\n", addr, a - addr, sym);
         }
      }
   }
}
#endif

#ifdef ia64_TARGET_ARCH
static unsigned int PLTSize(void);
#endif

/* -----------------------------------------------------------------------------
 * Load an obj (populate the global symbol table, but don't resolve yet)
 *
 * Returns: 1 if ok, 0 on error.
 */
HsInt
loadObj( char *path )
{
   ObjectCode* oc;
   struct stat st;
   int r, n;
#ifdef USE_MMAP
   int fd, pagesize;
   void *map_addr;
#else
   FILE *f;
#endif

   initLinker();

   /* fprintf(stderr, "loadObj %s\n", path ); */

   /* Check that we haven't already loaded this object.  Don't give up
      at this stage; ocGetNames_* will barf later. */
   {
       ObjectCode *o;
       int is_dup = 0;
       for (o = objects; o; o = o->next) {
          if (0 == strcmp(o->fileName, path))
             is_dup = 1;
       }
       if (is_dup) {
	 fprintf(stderr,
            "\n\n"
            "GHCi runtime linker: warning: looks like you're trying to load the\n"
            "same object file twice:\n"
            "   %s\n"
            "GHCi will continue, but a duplicate-symbol error may shortly follow.\n"
            "\n"
            , path);
       }
   }

   oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");

#  if defined(OBJFORMAT_ELF)
   oc->formatName = "ELF";
#  elif defined(OBJFORMAT_PEi386)
   oc->formatName = "PEi386";
#  elif defined(OBJFORMAT_MACHO)
   oc->formatName = "Mach-O";
#  else
   stgFree(oc);
   barf("loadObj: not implemented on this platform");
#  endif

   r = stat(path, &st);
   if (r == -1) { return 0; }

   /* sigh, strdup() isn't a POSIX function, so do it the long way */
   oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
   strcpy(oc->fileName, path);

   oc->fileSize          = st.st_size;
   oc->symbols           = NULL;
   oc->sections          = NULL;
   oc->lochash           = allocStrHashTable();
   oc->proddables        = NULL;

   /* chain it onto the list of objects */
   oc->next              = objects;
   objects               = oc;

#ifdef USE_MMAP
#define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))

   /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */

   fd = open(path, O_RDONLY);
   if (fd == -1)
      barf("loadObj: can't open `%s'", path);

   pagesize = getpagesize();

#ifdef ia64_TARGET_ARCH
   /* The PLT needs to be right before the object */
   n = ROUND_UP(PLTSize(), pagesize);
   oc->plt = mmap(NULL, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
   if (oc->plt == MAP_FAILED)
      barf("loadObj: can't allocate PLT");

   oc->pltIndex = 0;
   map_addr = oc->plt + n;
#endif

   n = ROUND_UP(oc->fileSize, pagesize);
   oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0);
   if (oc->image == MAP_FAILED)
      barf("loadObj: can't map `%s'", path);

   close(fd);

#else /* !USE_MMAP */

   oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");

   /* load the image into memory */
   f = fopen(path, "rb");
   if (!f)
       barf("loadObj: can't read `%s'", path);

   n = fread ( oc->image, 1, oc->fileSize, f );
   if (n != oc->fileSize)
      barf("loadObj: error whilst reading `%s'", path);

   fclose(f);

#endif /* USE_MMAP */

#  if defined(OBJFORMAT_MACHO)
   r = ocAllocateJumpIslands_MachO ( oc );
   if (!r) { return r; }
#endif

   /* verify the in-memory image */
#  if defined(OBJFORMAT_ELF)
   r = ocVerifyImage_ELF ( oc );
#  elif defined(OBJFORMAT_PEi386)
   r = ocVerifyImage_PEi386 ( oc );
#  elif defined(OBJFORMAT_MACHO)
   r = ocVerifyImage_MachO ( oc );
#  else
   barf("loadObj: no verify method");
#  endif
   if (!r) { return r; }

   /* build the symbol list for this image */
#  if defined(OBJFORMAT_ELF)
   r = ocGetNames_ELF ( oc );
#  elif defined(OBJFORMAT_PEi386)
   r = ocGetNames_PEi386 ( oc );
#  elif defined(OBJFORMAT_MACHO)
   r = ocGetNames_MachO ( oc );
#  else
   barf("loadObj: no getNames method");
#  endif
   if (!r) { return r; }

   /* loaded, but not resolved yet */
   oc->status = OBJECT_LOADED;

   return 1;
}

/* -----------------------------------------------------------------------------
 * resolve all the currently unlinked objects in memory
 *
 * Returns: 1 if ok, 0 on error.
 */
HsInt
resolveObjs( void )
{
    ObjectCode *oc;
    int r;

    initLinker();

    for (oc = objects; oc; oc = oc->next) {
	if (oc->status != OBJECT_RESOLVED) {
#           if defined(OBJFORMAT_ELF)
	    r = ocResolve_ELF ( oc );
#           elif defined(OBJFORMAT_PEi386)
	    r = ocResolve_PEi386 ( oc );
#           elif defined(OBJFORMAT_MACHO)
	    r = ocResolve_MachO ( oc );
#           else
	    barf("resolveObjs: not implemented on this platform");
#           endif
	    if (!r) { return r; }
	    oc->status = OBJECT_RESOLVED;
	}
    }
    return 1;
}

/* -----------------------------------------------------------------------------
 * delete an object from the pool
 */
HsInt
unloadObj( char *path )
{
    ObjectCode *oc, *prev;

    ASSERT(symhash != NULL);
    ASSERT(objects != NULL);

    initLinker(); 

    prev = NULL;
    for (oc = objects; oc; prev = oc, oc = oc->next) {
	if (!strcmp(oc->fileName,path)) {

	    /* Remove all the mappings for the symbols within this
	     * object..
	     */
	    {
                int i;
                for (i = 0; i < oc->n_symbols; i++) {
                   if (oc->symbols[i] != NULL) {
                       removeStrHashTable(symhash, oc->symbols[i], NULL);
                   }
                }
            }

	    if (prev == NULL) {
		objects = oc->next;
	    } else {
		prev->next = oc->next;
	    }

	    /* We're going to leave this in place, in case there are
	       any pointers from the heap into it: */
	    /* stgFree(oc->image); */
	    stgFree(oc->fileName);
	    stgFree(oc->symbols);
	    stgFree(oc->sections);
	    /* The local hash table should have been freed at the end
               of the ocResolve_ call on it. */
            ASSERT(oc->lochash == NULL);
	    stgFree(oc);
	    return 1;
	}
    }

    belch("unloadObj: can't find `%s' to unload", path);
    return 0;
}

/* -----------------------------------------------------------------------------
 * Sanity checking.  For each ObjectCode, maintain a list of address ranges
 * which may be prodded during relocation, and abort if we try and write
 * outside any of these.
 */
static void addProddableBlock ( ObjectCode* oc, void* start, int size )
{
   ProddableBlock* pb
      = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
   /* fprintf(stderr, "aPB %p %p %d\n", oc, start, size); */
   ASSERT(size > 0);
   pb->start      = start;
   pb->size       = size;
   pb->next       = oc->proddables;
   oc->proddables = pb;
}

static void checkProddableBlock ( ObjectCode* oc, void* addr )
{
   ProddableBlock* pb;
   for (pb = oc->proddables; pb != NULL; pb = pb->next) {
      char* s = (char*)(pb->start);
      char* e = s + pb->size - 1;
      char* a = (char*)addr;
      /* Assumes that the biggest fixup involves a 4-byte write.  This
         probably needs to be changed to 8 (ie, +7) on 64-bit
         plats. */
      if (a >= s && (a+3) <= e) return;
   }
   barf("checkProddableBlock: invalid fixup in runtime linker");
}

/* -----------------------------------------------------------------------------
 * Section management.
 */
static void addSection ( ObjectCode* oc, SectionKind kind,
                         void* start, void* end )
{
   Section* s   = stgMallocBytes(sizeof(Section), "addSection");
   s->start     = start;
   s->end       = end;
   s->kind      = kind;
   s->next      = oc->sections;
   oc->sections = s;
   /*
   fprintf(stderr, "addSection: %p-%p (size %d), kind %d\n",
                   start, ((char*)end)-1, end - start + 1, kind );
   */
}



/* --------------------------------------------------------------------------
 * PEi386 specifics (Win32 targets)
 * ------------------------------------------------------------------------*/

/* The information for this linker comes from
      Microsoft Portable Executable
      and Common Object File Format Specification
      revision 5.1 January 1998
   which SimonM says comes from the MS Developer Network CDs.

   It can be found there (on older CDs), but can also be found
   online at:

      http://www.microsoft.com/hwdev/hardware/PECOFF.asp

   (this is Rev 6.0 from February 1999).

   Things move, so if that fails, try searching for it via

      http://www.google.com/search?q=PE+COFF+specification

   The ultimate reference for the PE format is the Winnt.h
   header file that comes with the Platform SDKs; as always,
   implementations will drift wrt their documentation.

   A good background article on the PE format is Matt Pietrek's
   March 1994 article in Microsoft System Journal (MSJ)
   (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
   Win32 Portable Executable File Format." The info in there
   has recently been updated in a two part article in
   MSDN magazine, issues Feb and March 2002,
   "Inside Windows: An In-Depth Look into the Win32 Portable
   Executable File Format"

   John Levine's book "Linkers and Loaders" contains useful
   info on PE too.
*/


#if defined(OBJFORMAT_PEi386)



typedef unsigned char  UChar;
typedef unsigned short UInt16;
typedef unsigned int   UInt32;
typedef          int   Int32;


typedef
   struct {
      UInt16 Machine;
      UInt16 NumberOfSections;
      UInt32 TimeDateStamp;
      UInt32 PointerToSymbolTable;
      UInt32 NumberOfSymbols;
      UInt16 SizeOfOptionalHeader;
      UInt16 Characteristics;
   }
   COFF_header;

#define sizeof_COFF_header 20


typedef
   struct {
      UChar  Name[8];
      UInt32 VirtualSize;
      UInt32 VirtualAddress;
      UInt32 SizeOfRawData;
      UInt32 PointerToRawData;
      UInt32 PointerToRelocations;
      UInt32 PointerToLinenumbers;
      UInt16 NumberOfRelocations;
      UInt16 NumberOfLineNumbers;
      UInt32 Characteristics;
   }
   COFF_section;

#define sizeof_COFF_section 40


typedef
   struct {
      UChar  Name[8];
      UInt32 Value;
      UInt16 SectionNumber;
      UInt16 Type;
      UChar  StorageClass;
      UChar  NumberOfAuxSymbols;
   }
   COFF_symbol;

#define sizeof_COFF_symbol 18


typedef
   struct {
      UInt32 VirtualAddress;
      UInt32 SymbolTableIndex;
      UInt16 Type;
   }
   COFF_reloc;

#define sizeof_COFF_reloc 10


/* From PE spec doc, section 3.3.2 */
/* Note use of MYIMAGE_* since IMAGE_* are already defined in
   windows.h -- for the same purpose, but I want to know what I'm
   getting, here. */
#define MYIMAGE_FILE_RELOCS_STRIPPED     0x0001
#define MYIMAGE_FILE_EXECUTABLE_IMAGE    0x0002
#define MYIMAGE_FILE_DLL                 0x2000
#define MYIMAGE_FILE_SYSTEM              0x1000
#define MYIMAGE_FILE_BYTES_REVERSED_HI   0x8000
#define MYIMAGE_FILE_BYTES_REVERSED_LO   0x0080
#define MYIMAGE_FILE_32BIT_MACHINE       0x0100

/* From PE spec doc, section 5.4.2 and 5.4.4 */