X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Frts%2FLinker.c;h=a9faab5e2f6c6d4e616c12f0e3c445c6d93528fc;hb=04db0e9fa47ce4dfbcb73ec1752d94195f3b394e;hp=f507e479b3896caa86cdc8007801b5d5cc914e53;hpb=6317d554cb806e1eb03af252dc9744ae4dce10a1;p=ghc-hetmet.git diff --git a/ghc/rts/Linker.c b/ghc/rts/Linker.c index f507e47..a9faab5 100644 --- a/ghc/rts/Linker.c +++ b/ghc/rts/Linker.c @@ -10,7 +10,9 @@ #include "PosixSource.h" #endif -// Linux needs _GNU_SOURCE to get RTLD_DEFAULT from . +/* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from and + MREMAP_MAYMOVE from . + */ #ifdef __linux__ #define _GNU_SOURCE #endif @@ -24,6 +26,7 @@ #include "RtsUtils.h" #include "Schedule.h" #include "Storage.h" +#include "Sparks.h" #ifdef HAVE_SYS_TYPES_H #include @@ -36,9 +39,7 @@ #include #endif -#if defined(HAVE_FRAMEWORK_HASKELLSUPPORT) -#include -#elif defined(HAVE_DLFCN_H) +#if defined(HAVE_DLFCN_H) #include #endif @@ -58,12 +59,12 @@ #include #endif -#if defined(ia64_HOST_ARCH) || defined(openbsd_HOST_OS) || defined(linux_HOST_OS) +#if defined(ia64_HOST_ARCH) || defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS) #define USE_MMAP #include #include -#if defined(openbsd_HOST_OS) || defined(linux_HOST_OS) +#if defined(openbsd_HOST_OS) || defined(linux_HOST_OS) || defined(freebsd_HOST_OS) #ifdef HAVE_UNISTD_H #include #endif @@ -78,12 +79,14 @@ # include # include #elif defined(darwin_HOST_OS) -# include # define OBJFORMAT_MACHO # include # include # include # include +#if defined(powerpc_HOST_ARCH) +# include +#endif #endif /* Hash table mapping symbol names to Symbol */ @@ -104,13 +107,20 @@ 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 int machoGetMisalignment( FILE * ); +#ifdef powerpc_HOST_ARCH +static int ocAllocateJumpIslands_MachO ( ObjectCode* oc ); static void machoInitSymbolsWithoutUnderscore( void ); #endif +#endif + +#if defined(x86_64_HOST_ARCH) +static void*x86_64_high_symbol( char *lbl, void *addr ); +#endif /* ----------------------------------------------------------------------------- * Built-in symbols from the RTS @@ -123,19 +133,17 @@ typedef struct _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_HOST_OS) #define RTS_POSIX_ONLY_SYMBOLS \ + SymX(signal_handlers) \ SymX(stg_sig_install) \ Sym(nocldstop) #endif @@ -295,12 +303,27 @@ typedef struct _RtsSymbolVal { SymX(exp) \ SymX(log) \ SymX(sqrt) \ + SymX(powf) \ + SymX(tanhf) \ + SymX(coshf) \ + SymX(sinhf) \ + SymX(atanf) \ + SymX(acosf) \ + SymX(asinf) \ + SymX(tanf) \ + SymX(cosf) \ + SymX(sinf) \ + SymX(expf) \ + SymX(logf) \ + SymX(sqrtf) \ SymX(memcpy) \ - SymX(stg_InstallConsoleEvent) \ + SymX(rts_InstallConsoleEvent) \ + SymX(rts_ConsoleHandlerDone) \ Sym(mktime) \ Sym(_imp___timezone) \ Sym(_imp___tzname) \ Sym(_imp___iob) \ + Sym(_imp___osver) \ Sym(localtime) \ Sym(gmtime) \ Sym(opendir) \ @@ -310,19 +333,75 @@ typedef struct _RtsSymbolVal { Sym(closedir) #endif +#if defined(darwin_TARGET_OS) && HAVE_PRINTF_LDBLSTUB +#define RTS_DARWIN_ONLY_SYMBOLS \ + Sym(asprintf$LDBLStub) \ + Sym(err$LDBLStub) \ + Sym(errc$LDBLStub) \ + Sym(errx$LDBLStub) \ + Sym(fprintf$LDBLStub) \ + Sym(fscanf$LDBLStub) \ + Sym(fwprintf$LDBLStub) \ + Sym(fwscanf$LDBLStub) \ + Sym(printf$LDBLStub) \ + Sym(scanf$LDBLStub) \ + Sym(snprintf$LDBLStub) \ + Sym(sprintf$LDBLStub) \ + Sym(sscanf$LDBLStub) \ + Sym(strtold$LDBLStub) \ + Sym(swprintf$LDBLStub) \ + Sym(swscanf$LDBLStub) \ + Sym(syslog$LDBLStub) \ + Sym(vasprintf$LDBLStub) \ + Sym(verr$LDBLStub) \ + Sym(verrc$LDBLStub) \ + Sym(verrx$LDBLStub) \ + Sym(vfprintf$LDBLStub) \ + Sym(vfscanf$LDBLStub) \ + Sym(vfwprintf$LDBLStub) \ + Sym(vfwscanf$LDBLStub) \ + Sym(vprintf$LDBLStub) \ + Sym(vscanf$LDBLStub) \ + Sym(vsnprintf$LDBLStub) \ + Sym(vsprintf$LDBLStub) \ + Sym(vsscanf$LDBLStub) \ + Sym(vswprintf$LDBLStub) \ + Sym(vswscanf$LDBLStub) \ + Sym(vsyslog$LDBLStub) \ + Sym(vwarn$LDBLStub) \ + Sym(vwarnc$LDBLStub) \ + Sym(vwarnx$LDBLStub) \ + Sym(vwprintf$LDBLStub) \ + Sym(vwscanf$LDBLStub) \ + Sym(warn$LDBLStub) \ + Sym(warnc$LDBLStub) \ + Sym(warnx$LDBLStub) \ + Sym(wcstold$LDBLStub) \ + Sym(wprintf$LDBLStub) \ + Sym(wscanf$LDBLStub) +#else +#define RTS_DARWIN_ONLY_SYMBOLS +#endif + #ifndef SMP # define MAIN_CAP_SYM SymX(MainCapability) #else # define MAIN_CAP_SYM #endif +#if !defined(mingw32_HOST_OS) +#define RTS_USER_SIGNALS_SYMBOLS \ + SymX(setIOManagerPipe) +#else +#define RTS_USER_SIGNALS_SYMBOLS /* nothing */ +#endif + #ifdef TABLES_NEXT_TO_CODE #define RTS_RET_SYMBOLS /* nothing */ #else #define RTS_RET_SYMBOLS \ SymX(stg_enter_ret) \ SymX(stg_gc_fun_ret) \ - SymX(stg_ap_0_ret) \ SymX(stg_ap_v_ret) \ SymX(stg_ap_f_ret) \ SymX(stg_ap_d_ret) \ @@ -340,7 +419,6 @@ typedef struct _RtsSymbolVal { #endif #define RTS_SYMBOLS \ - Maybe_ForeignObj \ Maybe_Stable_Names \ Sym(StgReturn) \ SymX(stg_enter_info) \ @@ -409,6 +487,7 @@ typedef struct _RtsSymbolVal { SymX(delayzh_fast) \ SymX(deRefWeakzh_fast) \ SymX(deRefStablePtrzh_fast) \ + SymX(dirty_MUT_VAR) \ SymX(divExactIntegerzh_fast) \ SymX(divModIntegerzh_fast) \ SymX(forkzh_fast) \ @@ -423,7 +502,14 @@ typedef struct _RtsSymbolVal { SymX(genericRaise) \ SymX(getProgArgv) \ SymX(getStablePtr) \ - SymX(initLinker) \ + SymX(hs_init) \ + SymX(hs_exit) \ + SymX(hs_set_argv) \ + SymX(hs_add_root) \ + SymX(hs_perform_gc) \ + SymX(hs_free_stable_ptr) \ + SymX(hs_free_fun_ptr) \ + SymX(initLinker) \ SymX(int2Integerzh_fast) \ SymX(integer2Intzh_fast) \ SymX(integer2Wordzh_fast) \ @@ -454,6 +540,7 @@ typedef struct _RtsSymbolVal { SymX(newTVarzh_fast) \ SymX(atomicModifyMutVarzh_fast) \ SymX(newPinnedByteArrayzh_fast) \ + SymX(newSpark) \ SymX(orIntegerzh_fast) \ SymX(performGC) \ SymX(performMajorGC) \ @@ -511,7 +598,6 @@ typedef struct _RtsSymbolVal { SymX(rts_mkWord8) \ SymX(rts_unlock) \ SymX(rtsSupportsBoundThreads) \ - SymX(run_queue_hd) \ SymX(__hscore_get_saved_termios) \ SymX(__hscore_set_saved_termios) \ SymX(setProgArgv) \ @@ -521,15 +607,15 @@ typedef struct _RtsSymbolVal { 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_DIRTY_info) \ SymX(stg_MUT_ARR_PTRS_FROZEN_info) \ + SymX(stg_MUT_ARR_PTRS_FROZEN0_info) \ SymX(stg_WEAK_info) \ - SymX(stg_ap_0_info) \ SymX(stg_ap_v_info) \ SymX(stg_ap_f_info) \ SymX(stg_ap_d_info) \ @@ -544,6 +630,21 @@ typedef struct _RtsSymbolVal { SymX(stg_ap_pppp_info) \ SymX(stg_ap_ppppp_info) \ SymX(stg_ap_pppppp_info) \ + SymX(stg_ap_0_fast) \ + SymX(stg_ap_v_fast) \ + SymX(stg_ap_f_fast) \ + SymX(stg_ap_d_fast) \ + SymX(stg_ap_l_fast) \ + SymX(stg_ap_n_fast) \ + SymX(stg_ap_p_fast) \ + SymX(stg_ap_pv_fast) \ + SymX(stg_ap_pp_fast) \ + SymX(stg_ap_ppv_fast) \ + SymX(stg_ap_ppp_fast) \ + SymX(stg_ap_pppv_fast) \ + SymX(stg_ap_pppp_fast) \ + SymX(stg_ap_ppppp_fast) \ + SymX(stg_ap_pppppp_fast) \ SymX(stg_ap_1_upd_info) \ SymX(stg_ap_2_upd_info) \ SymX(stg_ap_3_upd_info) \ @@ -582,7 +683,21 @@ typedef struct _RtsSymbolVal { SymX(word2Integerzh_fast) \ SymX(writeTVarzh_fast) \ SymX(xorIntegerzh_fast) \ - SymX(yieldzh_fast) + SymX(yieldzh_fast) \ + SymX(stg_interp_constr_entry) \ + SymX(stg_interp_constr1_entry) \ + SymX(stg_interp_constr2_entry) \ + SymX(stg_interp_constr3_entry) \ + SymX(stg_interp_constr4_entry) \ + SymX(stg_interp_constr5_entry) \ + SymX(stg_interp_constr6_entry) \ + SymX(stg_interp_constr7_entry) \ + SymX(stg_interp_constr8_entry) \ + SymX(stgMallocBytesRWX) \ + SymX(getAllocations) \ + SymX(revertCAFs) \ + SymX(RtsFlags) \ + RTS_USER_SIGNALS_SYMBOLS #ifdef SUPPORT_LONG_LONGS #define RTS_LONG_LONG_SYMS \ @@ -616,7 +731,7 @@ typedef struct _RtsSymbolVal { #define RTS_LIBGCC_SYMBOLS #endif -#ifdef darwin_HOST_OS +#if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH) // Symbols that don't have a leading underscore // on Mac OS X. They have to receive special treatment, // see machoInitSymbolsWithoutUnderscore() @@ -635,6 +750,7 @@ RTS_LONG_LONG_SYMS RTS_POSIX_ONLY_SYMBOLS RTS_MINGW_ONLY_SYMBOLS RTS_CYGWIN_ONLY_SYMBOLS +RTS_DARWIN_ONLY_SYMBOLS RTS_LIBGCC_SYMBOLS #undef Sym #undef SymX @@ -664,6 +780,12 @@ static RtsSymbolVal rtsSyms[] = { RTS_MINGW_ONLY_SYMBOLS RTS_CYGWIN_ONLY_SYMBOLS RTS_LIBGCC_SYMBOLS +#if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH) + // dyld stub code contains references to this, + // but it should never be called because we treat + // lazy pointers as nonlazy. + { "dyld_stub_binding_helper", (void*)0xDEADBEEF }, +#endif { 0, 0 } /* sentinel */ }; @@ -736,7 +858,7 @@ initLinker( void ) ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, sym->lbl, sym->addr); } -# if defined(OBJFORMAT_MACHO) +# if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH) machoInitSymbolsWithoutUnderscore(); # endif @@ -748,7 +870,7 @@ initLinker( void ) # if defined(openbsd_HOST_OS) dl_libc_handle = dlopen("libc.so", RTLD_LAZY); # endif -# endif // RTLD_DEFAULT +# endif /* RTLD_DEFAULT */ # endif } @@ -881,6 +1003,16 @@ lookupSymbol( char *lbl ) # if defined(openbsd_HOST_OS) val = dlsym(dl_prog_handle, lbl); return (val != NULL) ? val : dlsym(dl_libc_handle,lbl); +# elif defined(x86_64_HOST_ARCH) + val = dlsym(dl_prog_handle, lbl); + if (val >= (void *)0x80000000) { + void *new_val; + new_val = x86_64_high_symbol(lbl, val); + IF_DEBUG(linker,debugBelch("lookupSymbol: relocating out of range symbol: %s = %p, now %p\n", lbl, val, new_val)); + return new_val; + } else { + return val; + } # else /* not openbsd */ return dlsym(dl_prog_handle, lbl); # endif @@ -974,7 +1106,7 @@ void ghci_enquire ( char* addr ) // debugBelch("ghci_enquire: can't find %s\n", sym); } else if (addr-DELTA <= a && a <= addr+DELTA) { - debugBelch("%p + %3d == `%s'\n", addr, a - addr, sym); + debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym); } } } @@ -1001,8 +1133,8 @@ loadObj( char *path ) void *map_addr = NULL; #else FILE *f; + int misalignment; #endif - initLinker(); /* debugBelch("loadObj %s\n", path ); */ @@ -1086,7 +1218,19 @@ loadObj( char *path ) #endif n = ROUND_UP(oc->fileSize, pagesize); - oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); + + /* Link objects into the lower 2Gb on x86_64. GHC assumes the + * small memory model on this architecture (see gcc docs, + * -mcmodel=small). + */ +#ifdef x86_64_HOST_ARCH +#define EXTRA_MAP_FLAGS MAP_32BIT +#else +#define EXTRA_MAP_FLAGS 0 +#endif + + oc->image = mmap(map_addr, n, PROT_EXEC|PROT_READ|PROT_WRITE, + MAP_PRIVATE|EXTRA_MAP_FLAGS, fd, 0); if (oc->image == MAP_FAILED) barf("loadObj: can't map `%s'", path); @@ -1094,13 +1238,30 @@ loadObj( char *path ) #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); +#ifdef darwin_HOST_OS + // In a Mach-O .o file, all sections can and will be misaligned + // if the total size of the headers is not a multiple of the + // desired alignment. This is fine for .o files that only serve + // as input for the static linker, but it's not fine for us, + // as SSE (used by gcc for floating point) and Altivec require + // 16-byte alignment. + // We calculate the correct alignment from the header before + // reading the file, and then we misalign oc->image on purpose so + // that the actual sections end up aligned again. + misalignment = machoGetMisalignment(f); + oc->misalignment = misalignment; +#else + misalignment = 0; +#endif + + oc->image = stgMallocBytes(oc->fileSize + misalignment, "loadObj(image)"); + oc->image += misalignment; + n = fread ( oc->image, 1, oc->fileSize, f ); if (n != oc->fileSize) barf("loadObj: error whilst reading `%s'", path); @@ -1109,7 +1270,7 @@ loadObj( char *path ) #endif /* USE_MMAP */ -# if defined(OBJFORMAT_MACHO) +# if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH) r = ocAllocateJumpIslands_MachO ( oc ); if (!r) { return r; } # elif defined(OBJFORMAT_ELF) && defined(powerpc_HOST_ARCH) @@ -1306,21 +1467,73 @@ static void addSection ( ObjectCode* oc, SectionKind kind, static int ocAllocateJumpIslands( ObjectCode* oc, int count, int first ) { +#ifdef USE_MMAP + int pagesize, n, m; +#endif int aligned; + int misalignment = 0; +#if darwin_HOST_OS + misalignment = oc->misalignment; +#endif if( count > 0 ) { -#ifdef USE_MMAP - #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined -#endif // round up to the nearest 4 aligned = (oc->fileSize + 3) & ~3; +#ifdef USE_MMAP + #ifndef linux_HOST_OS /* mremap is a linux extension */ + #error ocAllocateJumpIslands doesnt want USE_MMAP to be defined + #endif + + pagesize = getpagesize(); + n = ROUND_UP( oc->fileSize, pagesize ); + m = ROUND_UP( aligned + sizeof (ppcJumpIsland) * count, pagesize ); + + /* If we have a half-page-size file and map one page of it then + * the part of the page after the size of the file remains accessible. + * If, however, we map in 2 pages, the 2nd page is not accessible + * and will give a "Bus Error" on access. To get around this, we check + * if we need any extra pages for the jump islands and map them in + * anonymously. We must check that we actually require extra pages + * otherwise the attempt to mmap 0 pages of anonymous memory will + * fail -EINVAL. + */ + + if( m > n ) + { + /* The effect of this mremap() call is only the ensure that we have + * a sufficient number of virtually contiguous pages. As returned from + * mremap, the pages past the end of the file are not backed. We give + * them a backing by using MAP_FIXED to map in anonymous pages. + */ + oc->image = mremap( oc->image, n, m, MREMAP_MAYMOVE ); + + if( oc->image == MAP_FAILED ) + { + errorBelch( "Unable to mremap for Jump Islands\n" ); + return 0; + } + + if( mmap( oc->image + n, m - n, PROT_READ | PROT_WRITE | PROT_EXEC, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, 0, 0 ) == MAP_FAILED ) + { + errorBelch( "Unable to mmap( MAP_FIXED ) for Jump Islands\n" ); + return 0; + } + } + +#else + oc->image -= misalignment; oc->image = stgReallocBytes( oc->image, - aligned + sizeof( ppcJumpIsland ) * count, + misalignment + + aligned + sizeof (ppcJumpIsland) * count, "ocAllocateJumpIslands" ); - oc->jump_islands = (ppcJumpIsland *) (((char *) oc->image) + aligned); - memset( oc->jump_islands, 0, sizeof( ppcJumpIsland ) * count ); + oc->image += misalignment; +#endif /* USE_MMAP */ + + oc->jump_islands = (ppcJumpIsland *) (oc->image + aligned); + memset( oc->jump_islands, 0, sizeof (ppcJumpIsland) * count ); } else oc->jump_islands = NULL; @@ -1670,24 +1883,25 @@ ocVerifyImage_PEi386 ( ObjectCode* oc ) + hdr->NumberOfSymbols * sizeof_COFF_symbol; if (hdr->Machine != 0x14c) { - errorBelch("Not x86 PEi386"); + errorBelch("%s: Not x86 PEi386", oc->fileName); return 0; } if (hdr->SizeOfOptionalHeader != 0) { - errorBelch("PEi386 with nonempty optional header"); + errorBelch("%s: PEi386 with nonempty optional header", oc->fileName); return 0; } if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */ (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) || (hdr->Characteristics & MYIMAGE_FILE_DLL) || (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) { - errorBelch("Not a PEi386 object file"); + errorBelch("%s: Not a PEi386 object file", oc->fileName); return 0; } if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI) /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) { - errorBelch("Invalid PEi386 word size or endiannness: %d", - (int)(hdr->Characteristics)); + errorBelch("%s: Invalid PEi386 word size or endiannness: %d", + oc->fileName, + (int)(hdr->Characteristics)); return 0; } /* If the string table size is way crazy, this might indicate that @@ -1862,19 +2076,37 @@ ocGetNames_PEi386 ( ObjectCode* oc ) /* Allocate space for any (local, anonymous) .bss sections. */ for (i = 0; i < hdr->NumberOfSections; i++) { + UInt32 bss_sz; UChar* zspace; COFF_section* sectab_i = (COFF_section*) myindex ( sizeof_COFF_section, sectab, i ); if (0 != strcmp(sectab_i->Name, ".bss")) continue; - if (sectab_i->VirtualSize == 0) continue; + /* sof 10/05: the PE spec text isn't too clear regarding what + * the SizeOfRawData field is supposed to hold for object + * file sections containing just uninitialized data -- for executables, + * it is supposed to be zero; unclear what it's supposed to be + * for object files. However, VirtualSize is guaranteed to be + * zero for object files, which definitely suggests that SizeOfRawData + * will be non-zero (where else would the size of this .bss section be + * stored?) Looking at the COFF_section info for incoming object files, + * this certainly appears to be the case. + * + * => I suspect we've been incorrectly handling .bss sections in (relocatable) + * object files up until now. This turned out to bite us with ghc-6.4.1's use + * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static' + * variable decls into to the .bss section. (The specific function in Q which + * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath()) + */ + if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue; /* This is a non-empty .bss section. Allocate zeroed space for it, and set its PointerToRawData field such that oc->image + PointerToRawData == addr_of_zeroed_space. */ - zspace = stgCallocBytes(1, sectab_i->VirtualSize, - "ocGetNames_PEi386(anonymous bss)"); + bss_sz = sectab_i->VirtualSize; + if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; } + zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)"); sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image)); - addProddableBlock(oc, zspace, sectab_i->VirtualSize); + addProddableBlock(oc, zspace, bss_sz); /* debugBelch("BSS anon section at 0x%x\n", zspace); */ } @@ -1903,6 +2135,7 @@ ocGetNames_PEi386 ( ObjectCode* oc ) # endif if (0==strcmp(".text",sectab_i->Name) || + 0==strcmp(".rdata",sectab_i->Name)|| 0==strcmp(".rodata",sectab_i->Name)) kind = SECTIONKIND_CODE_OR_RODATA; if (0==strcmp(".data",sectab_i->Name) || @@ -1921,8 +2154,10 @@ ocGetNames_PEi386 ( ObjectCode* oc ) information. */ && 0 != strcmp(".stab", sectab_i->Name) && 0 != strcmp(".stabstr", sectab_i->Name) + /* ignore constructor section for now */ + && 0 != strcmp(".ctors", sectab_i->Name) ) { - errorBelch("Unknown PEi386 section name `%s'", sectab_i->Name); + errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName); return 0; } @@ -2064,7 +2299,8 @@ ocResolve_PEi386 ( ObjectCode* oc ) /* Ignore sections called which contain stabs debugging information. */ if (0 == strcmp(".stab", sectab_i->Name) - || 0 == strcmp(".stabstr", sectab_i->Name)) + || 0 == strcmp(".stabstr", sectab_i->Name) + || 0 == strcmp(".ctors", sectab_i->Name)) continue; if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) { @@ -2081,8 +2317,14 @@ ocResolve_PEi386 ( ObjectCode* oc ) COFF_reloc* rel = (COFF_reloc*) myindex ( sizeof_COFF_reloc, reltab, 0 ); noRelocs = rel->VirtualAddress; + + /* 10/05: we now assume (and check for) a GNU ld that is capable + * of handling object files with (>2^16) of relocs. + */ +#if 0 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n", noRelocs); +#endif j = 1; } else { noRelocs = sectab_i->NumberOfRelocations; @@ -2143,7 +2385,7 @@ ocResolve_PEi386 ( ObjectCode* oc ) /* Newline first because the interactive linker has printed "linking..." */ errorBelch("\n%s: unknown symbol `%s'", oc->fileName, symbol); return 0; - foundit: + foundit:; } checkProddableBlock(oc, pP); switch (reltab_j->Type) { @@ -2161,9 +2403,22 @@ ocResolve_PEi386 ( ObjectCode* oc ) -- hence the constant 4. Also I don't know if A should be added, but so far it has always been zero. + + SOF 05/2005: 'A' (old contents of *pP) have been observed + to contain values other than zero (the 'wx' object file + that came with wxhaskell-0.9.4; dunno how it was compiled..). + So, add displacement to old value instead of asserting + A to be zero. Fixes wxhaskell-related crashes, and no other + ill effects have been observed. + + Update: the reason why we're seeing these more elaborate + relocations is due to a switch in how the NCG compiles SRTs + and offsets to them from info tables. SRTs live in .(ro)data, + while info tables live in .text, causing GAS to emit REL32/DISP32 + relocations with non-zero values. Adding the displacement is + the right thing to do. */ - ASSERT(A==0); - *pP = S - ((UInt32)pP) - 4; + *pP = S - ((UInt32)pP) - 4 + A; break; default: debugBelch("%s: unhandled PEi386 relocation type %d", @@ -2379,6 +2634,64 @@ PLTSize(void) #endif +#if x86_64_HOST_ARCH +// On x86_64, 32-bit relocations are often used, which requires that +// we can resolve a symbol to a 32-bit offset. However, shared +// libraries are placed outside the 2Gb area, which leaves us with a +// problem when we need to give a 32-bit offset to a symbol in a +// shared library. +// +// For a function symbol, we can allocate a bounce sequence inside the +// 2Gb area and resolve the symbol to this. The bounce sequence is +// simply a long jump instruction to the real location of the symbol. +// +// For data references, we're screwed. +// +typedef struct { + unsigned char jmp[8]; /* 6 byte instruction: jmpq *0x00000002(%rip) */ + void *addr; +} x86_64_bounce; + +#define X86_64_BB_SIZE 1024 + +static x86_64_bounce *x86_64_bounce_buffer = NULL; +static nat x86_64_bb_next_off; + +static void* +x86_64_high_symbol( char *lbl, void *addr ) +{ + x86_64_bounce *bounce; + + if ( x86_64_bounce_buffer == NULL || + x86_64_bb_next_off >= X86_64_BB_SIZE ) { + x86_64_bounce_buffer = + mmap(NULL, X86_64_BB_SIZE * sizeof(x86_64_bounce), + PROT_EXEC|PROT_READ|PROT_WRITE, + MAP_PRIVATE|MAP_32BIT|MAP_ANONYMOUS, -1, 0); + if (x86_64_bounce_buffer == MAP_FAILED) { + barf("x86_64_high_symbol: mmap failed"); + } + x86_64_bb_next_off = 0; + } + bounce = &x86_64_bounce_buffer[x86_64_bb_next_off]; + bounce->jmp[0] = 0xff; + bounce->jmp[1] = 0x25; + bounce->jmp[2] = 0x02; + bounce->jmp[3] = 0x00; + bounce->jmp[4] = 0x00; + bounce->jmp[5] = 0x00; + bounce->addr = addr; + x86_64_bb_next_off++; + + IF_DEBUG(linker, debugBelch("x86_64: allocated bounce entry for %s->%p at %p\n", + lbl, addr, bounce)); + + insertStrHashTable(symhash, lbl, bounce); + return bounce; +} +#endif + + /* * Generic ELF functions */ @@ -2476,14 +2789,17 @@ ocVerifyImage_ELF ( ObjectCode* oc ) case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break; #endif case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break; +#ifdef EM_X86_64 + case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break; +#endif default: IF_DEBUG(linker,debugBelch( "unknown" )); errorBelch("%s: unknown architecture", oc->fileName); return 0; } IF_DEBUG(linker,debugBelch( - "\nSection header table: start %d, n_entries %d, ent_size %d\n", - ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize )); + "\nSection header table: start %ld, n_entries %d, ent_size %d\n", + (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize )); ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr)); @@ -2548,9 +2864,9 @@ ocVerifyImage_ELF ( ObjectCode* oc ) nsymtabs++; stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset); nent = shdr[i].sh_size / sizeof(Elf_Sym); - IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%d rem)\n", + IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n", nent, - shdr[i].sh_size % sizeof(Elf_Sym) + (long)shdr[i].sh_size % sizeof(Elf_Sym) )); if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) { errorBelch("%s: non-integral number of symbol table entries", oc->fileName); @@ -2859,8 +3175,8 @@ do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC, case R_386_PC32: *pP = value - P; break; # endif default: - errorBelch("%s: unhandled ELF relocation(Rel) type %d\n", - oc->fileName, ELF_R_TYPE(info)); + errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n", + oc->fileName, (lnat)ELF_R_TYPE(info)); return 0; } @@ -2876,7 +3192,7 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, Elf_Sym* stab, char* strtab ) { int j; - char *symbol; + char *symbol = NULL; Elf_Addr targ; Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset); int nent = shdr[shnum].sh_size / sizeof(Elf_Rela); @@ -2889,7 +3205,7 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, target_shndx, symtab_shndx )); for (j = 0; j < nent; j++) { -#if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) +#if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH) /* This #ifdef only serves to avoid unused-var warnings. */ Elf_Addr offset = rtab[j].r_offset; Elf_Addr P = targ + offset; @@ -3065,9 +3381,43 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, | (delta & 0x3fffffc); break; # endif + +#if x86_64_HOST_ARCH + case R_X86_64_64: + *(Elf64_Xword *)P = value; + break; + + case R_X86_64_PC32: + { + StgInt64 off = value - P; + if (off >= 0x7fffffffL || off < -0x80000000L) { + barf("R_X86_64_PC32 relocation out of range: %s = %p", + symbol, off); + } + *(Elf64_Word *)P = (Elf64_Word)off; + break; + } + + case R_X86_64_32: + if (value >= 0x7fffffffL) { + barf("R_X86_64_32 relocation out of range: %s = %p\n", + symbol, value); + } + *(Elf64_Word *)P = (Elf64_Word)value; + break; + + case R_X86_64_32S: + if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) { + barf("R_X86_64_32S relocation out of range: %s = %p\n", + symbol, value); + } + *(Elf64_Sword *)P = (Elf64_Sword)value; + break; +#endif + default: - errorBelch("%s: unhandled ELF relocation(RelA) type %d\n", - oc->fileName, ELF_R_TYPE(info)); + errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n", + oc->fileName, (lnat)ELF_R_TYPE(info)); return 0; } @@ -3136,7 +3486,7 @@ ia64_extract_instruction(Elf64_Xword *target) { Elf64_Xword w1, w2; int slot = (Elf_Addr)target & 3; - (Elf_Addr)target &= ~3; + target = (Elf_Addr)target & ~3; w1 = *target; w2 = *(target+1); @@ -3158,7 +3508,7 @@ static void ia64_deposit_instruction(Elf64_Xword *target, Elf64_Xword value) { int slot = (Elf_Addr)target & 3; - (Elf_Addr)target &= ~3; + target = (Elf_Addr)target & ~3; switch (slot) { @@ -3261,7 +3611,7 @@ static int ocAllocateJumpIslands_ELF( ObjectCode *oc ) #if defined(OBJFORMAT_MACHO) /* - Support for MachO linking on Darwin/MacOS X on PowerPC chips + Support for MachO linking on Darwin/MacOS X by Wolfgang Thaller (wolfgang.thaller@gmx.net) I hereby formally apologize for the hackish nature of this code. @@ -3270,6 +3620,7 @@ static int ocAllocateJumpIslands_ELF( ObjectCode *oc ) *) add still more sanity checks. */ +#ifdef powerpc_HOST_ARCH static int ocAllocateJumpIslands_MachO(ObjectCode* oc) { struct mach_header *header = (struct mach_header *) oc->image; @@ -3284,7 +3635,7 @@ static int ocAllocateJumpIslands_MachO(ObjectCode* oc) // symbol, so we don't have to allocate too many // jump islands. struct symtab_command *symLC = (struct symtab_command *) lc; - int min = symLC->nsyms, max = 0; + unsigned min = symLC->nsyms, max = 0; struct nlist *nlist = symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff) : NULL; @@ -3314,8 +3665,9 @@ static int ocAllocateJumpIslands_MachO(ObjectCode* oc) } return ocAllocateJumpIslands(oc,0,0); } +#endif -static int ocVerifyImage_MachO(ObjectCode* oc) +static int ocVerifyImage_MachO(ObjectCode* oc STG_UNUSED) { // FIXME: do some verifying here return 1; @@ -3358,7 +3710,7 @@ static int resolveImports( return 1; } -static char* relocateAddress( +static unsigned long relocateAddress( ObjectCode* oc, int nSections, struct section* sections, @@ -3370,12 +3722,13 @@ static char* relocateAddress( if(sections[i].addr <= address && address < sections[i].addr + sections[i].size) { - return oc->image + sections[i].offset + address - sections[i].addr; + return (unsigned long)oc->image + + sections[i].offset + address - sections[i].addr; } } barf("Invalid Mach-O file:" "Address out of bounds while relocating object file"); - return NULL; + return 0; } static int relocateSection( @@ -3391,6 +3744,10 @@ static int relocateSection( return 1; else if(!strcmp(sect->sectname,"__nl_symbol_ptr")) return 1; + else if(!strcmp(sect->sectname,"__la_sym_ptr2")) + return 1; + else if(!strcmp(sect->sectname,"__la_sym_ptr3")) + return 1; n = sect->nreloc; relocs = (struct relocation_info*) (image + sect->reloff); @@ -3410,6 +3767,15 @@ static int relocateSection( unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address); checkProddableBlock(oc,wordPtr); + // Note on relocation types: + // i386 uses the GENERIC_RELOC_* types, + // while ppc uses special PPC_RELOC_* types. + // *_RELOC_VANILLA and *_RELOC_PAIR have the same value + // in both cases, all others are different. + // Therefore, we use GENERIC_RELOC_VANILLA + // and GENERIC_RELOC_PAIR instead of the PPC variants, + // and use #ifdefs for the other types. + // Step 1: Figure out what the relocated value should be if(scat->r_type == GENERIC_RELOC_VANILLA) { @@ -3420,23 +3786,28 @@ static int relocateSection( scat->r_value) - scat->r_value; } +#ifdef powerpc_HOST_ARCH else if(scat->r_type == PPC_RELOC_SECTDIFF || scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HA16_SECTDIFF) +#else + else if(scat->r_type == GENERIC_RELOC_SECTDIFF) +#endif { struct scattered_relocation_info *pair = (struct scattered_relocation_info*) &relocs[i+1]; - if(!pair->r_scattered || pair->r_type != PPC_RELOC_PAIR) + if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR) barf("Invalid Mach-O file: " - "PPC_RELOC_*_SECTDIFF not followed by PPC_RELOC_PAIR"); + "RELOC_*_SECTDIFF not followed by RELOC_PAIR"); word = (unsigned long) (relocateAddress(oc, nSections, sections, scat->r_value) - relocateAddress(oc, nSections, sections, pair->r_value)); i++; } +#ifdef powerpc_HOST_ARCH else if(scat->r_type == PPC_RELOC_HI16 || scat->r_type == PPC_RELOC_LO16 || scat->r_type == PPC_RELOC_HA16 @@ -3475,14 +3846,21 @@ static int relocateSection( i++; } + #endif else continue; // ignore the others +#ifdef powerpc_HOST_ARCH if(scat->r_type == GENERIC_RELOC_VANILLA || scat->r_type == PPC_RELOC_SECTDIFF) +#else + if(scat->r_type == GENERIC_RELOC_VANILLA + || scat->r_type == GENERIC_RELOC_SECTDIFF) +#endif { *wordPtr = word; } +#ifdef powerpc_HOST_ARCH else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16) { ((unsigned short*) wordPtr)[1] = word & 0xFFFF; @@ -3496,6 +3874,7 @@ static int relocateSection( ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF) + ((word & (1<<15)) ? 1 : 0); } +#endif } } @@ -3510,8 +3889,12 @@ static int relocateSection( if(reloc->r_length == 2) { unsigned long word = 0; +#ifdef powerpc_HOST_ARCH unsigned long jumpIsland = 0; - long offsetToJumpIsland; + long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value + // to avoid warning and to catch + // bugs. +#endif unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address); checkProddableBlock(oc,wordPtr); @@ -3520,6 +3903,7 @@ static int relocateSection( { word = *wordPtr; } +#ifdef powerpc_HOST_ARCH else if(reloc->r_type == PPC_RELOC_LO16) { word = ((unsigned short*) wordPtr)[1]; @@ -3540,7 +3924,7 @@ static int relocateSection( word = *wordPtr; word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0); } - +#endif if(!reloc->r_extern) { @@ -3555,7 +3939,7 @@ static int relocateSection( { struct nlist *symbol = &nlist[reloc->r_symbolnum]; char *nm = image + symLC->stroff + symbol->n_un.n_strx; - unsigned long symbolAddress = (unsigned long) (lookupSymbol(nm)); + void *symbolAddress = lookupSymbol(nm); if(!symbolAddress) { errorBelch("\nunknown symbol `%s'", nm); @@ -3564,21 +3948,23 @@ static int relocateSection( if(reloc->r_pcrel) { +#ifdef powerpc_HOST_ARCH // In the .o file, this should be a relative jump to NULL - // and we'll change it to a jump to a relative jump to the symbol + // and we'll change it to a relative jump to the symbol ASSERT(-word == reloc->r_address); - word = symbolAddress; - jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,word); - word -= ((long)image) + sect->offset + reloc->r_address; + jumpIsland = makeJumpIsland(oc,reloc->r_symbolnum,(unsigned long) symbolAddress); if(jumpIsland != 0) { - offsetToJumpIsland = jumpIsland - - (((long)image) + sect->offset + reloc->r_address); + offsetToJumpIsland = word + jumpIsland + - (((long)image) + sect->offset - sect->addr); } +#endif + word += (unsigned long) symbolAddress + - (((long)image) + sect->offset - sect->addr); } else { - word += symbolAddress; + word += (unsigned long) symbolAddress; } } @@ -3587,6 +3973,7 @@ static int relocateSection( *wordPtr = word; continue; } +#ifdef powerpc_HOST_ARCH else if(reloc->r_type == PPC_RELOC_LO16) { ((unsigned short*) wordPtr)[1] = word & 0xFFFF; @@ -3623,7 +4010,8 @@ static int relocateSection( *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC); continue; } - } +#endif + } barf("\nunknown relocation %d",reloc->r_type); return 0; } @@ -3636,7 +4024,7 @@ static int ocGetNames_MachO(ObjectCode* oc) char *image = (char*) oc->image; struct mach_header *header = (struct mach_header*) image; struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header)); - unsigned i,curSymbol; + unsigned i,curSymbol = 0; struct segment_command *segLC = NULL; struct section *sections; struct symtab_command *symLC = NULL; @@ -3718,7 +4106,6 @@ static int ocGetNames_MachO(ObjectCode* oc) if(symLC) { - curSymbol = 0; for(i=0;insyms;i++) { if(nlist[i].n_type & N_STAB) @@ -3806,6 +4193,10 @@ static int ocResolve_MachO(ObjectCode* oc) la_ptrs = §ions[i]; else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")) nl_ptrs = §ions[i]; + else if(!strcmp(sections[i].sectname,"__la_sym_ptr2")) + la_ptrs = §ions[i]; + else if(!strcmp(sections[i].sectname,"__la_sym_ptr3")) + la_ptrs = §ions[i]; } if(dsymLC) @@ -3838,6 +4229,7 @@ static int ocResolve_MachO(ObjectCode* oc) return 1; } +#ifdef powerpc_HOST_ARCH /* * The Mach-O object format uses leading underscores. But not everywhere. * There is a small number of runtime support functions defined in @@ -3851,7 +4243,7 @@ static void machoInitSymbolsWithoutUnderscore() { extern void* symbolsWithoutUnderscore[]; void **p = symbolsWithoutUnderscore; - __asm__ volatile(".data\n_symbolsWithoutUnderscore:"); + __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:"); #undef Sym #define Sym(x) \ @@ -3870,3 +4262,26 @@ static void machoInitSymbolsWithoutUnderscore() #undef Sym } #endif + +/* + * Figure out by how much to shift the entire Mach-O file in memory + * when loading so that its single segment ends up 16-byte-aligned + */ +static int machoGetMisalignment( FILE * f ) +{ + struct mach_header header; + int misalignment; + + fread(&header, sizeof(header), 1, f); + rewind(f); + + if(header.magic != MH_MAGIC) + return 0; + + misalignment = (header.sizeofcmds + sizeof(header)) + & 0xF; + + return misalignment ? (16 - misalignment) : 0; +} + +#endif