diff options
author | Pepe Iborra <pepeiborra@fb.com> | 2020-11-04 13:56:54 -0800 |
---|---|---|
committer | GHC GitLab CI <ghc-ci@gitlab-haskell.org> | 2020-11-24 01:17:25 +0000 |
commit | 552714a205c4b4f3346328fd12d2e5c64ba817a4 (patch) | |
tree | 687dd4dd24ff6d8f0112cd4d1e13f2d1b72d3660 | |
parent | eecf9f2726fb2508314a42860915edd13c752b34 (diff) | |
download | haskell-552714a205c4b4f3346328fd12d2e5c64ba817a4.tar.gz |
Revert "Revert "Add a RTS option -xp to load PIC object anywhere in address space""
This reverts commit 82a716431cc680392e332bc2b1a1fd0d7faa4cd8.
-rw-r--r-- | docs/users_guide/8.8.1-notes.rst | 5 | ||||
-rw-r--r-- | docs/users_guide/runtime_control.rst | 28 | ||||
-rw-r--r-- | includes/rts/Flags.h | 12 | ||||
-rw-r--r-- | libraries/base/GHC/RTS/Flags.hsc | 3 | ||||
-rw-r--r-- | rts/Linker.c | 79 | ||||
-rw-r--r-- | rts/RtsFlags.c | 11 | ||||
-rw-r--r-- | rts/linker/Elf.c | 25 | ||||
-rw-r--r-- | rts/linker/MachO.c | 21 | ||||
-rw-r--r-- | rts/linker/SymbolExtras.c | 61 |
9 files changed, 152 insertions, 93 deletions
diff --git a/docs/users_guide/8.8.1-notes.rst b/docs/users_guide/8.8.1-notes.rst index 6c10205b25..3b5ec80240 100644 --- a/docs/users_guide/8.8.1-notes.rst +++ b/docs/users_guide/8.8.1-notes.rst @@ -152,6 +152,11 @@ Runtime system - The output filename used for :ref:`eventlog output <rts-eventlog>` can now be specified with the :rts-flag:`-ol` flag. +- A new flag ``-xp`` is added on x86_64. When it is passed, the runtime linker + can load object files compiled with ``-fPIC -fexternal-dynamic-refs`` + anywhere in the address space. This used to be restricted to the low 2Gb. + + Template Haskell ~~~~~~~~~~~~~~~~ diff --git a/docs/users_guide/runtime_control.rst b/docs/users_guide/runtime_control.rst index a5c2f1474f..0048a51837 100644 --- a/docs/users_guide/runtime_control.rst +++ b/docs/users_guide/runtime_control.rst @@ -241,6 +241,28 @@ Miscellaneous RTS options crashes if exception handling are enabled. In order to get more information in compiled executables, C code or DLLs symbols need to be available. + +.. rts-flag:: -xp + + On 64-bit machines, the runtime linker usually needs to map object code + into the low 2Gb of the address space, due to the x86_64 small memory model + where most symbol references are 32 bits. The problem is that this 2Gb of + address space can fill up, especially if you're loading a very large number + of object files into GHCi. + + This flag offers a workaround, albeit a slightly convoluted one. To be able + to load an object file outside of the low 2Gb, the object code needs to be + compiled with ``-fPIC -fexternal-dynamic-refs``. When the ``+RTS -xp`` flag + is passed, the linker will assume that all object files were compiled with + ``-fPIC -fexternal-dynamic-refs`` and load them anywhere in the address + space. It's up to you to arrange that the object files you load (including + all packages) were compiled in the right way. If this is not the case for + an object, the linker will probably fail with an error message when the + problem is detected. + + On some platforms where PIC is always the case, e.g. x86_64 MacOS X, this + flag is enabled by default. + .. rts-flag:: -xm ⟨address⟩ .. index:: @@ -250,8 +272,10 @@ Miscellaneous RTS options This option is for working around memory allocation problems only. Do not use unless GHCi fails with a message like - “\ ``failed to mmap() memory below 2Gb``\ ”. If you need to use this - option to get GHCi working on your machine, please file a bug. + “\ ``failed to mmap() memory below 2Gb``\ ”. Consider recompiling + the objects with ``-fPIC -fexternal-dynamic-refs`` and using the + ``-xp`` flag instead. If you need to use this option to get GHCi + working on your machine, please file a bug. On 64-bit machines, the RTS needs to allocate memory in the low 2Gb of the address space. Support for this across different operating diff --git a/includes/rts/Flags.h b/includes/rts/Flags.h index 63450d5ece..3582e6abde 100644 --- a/includes/rts/Flags.h +++ b/includes/rts/Flags.h @@ -188,6 +188,17 @@ typedef struct _CONCURRENT_FLAGS { */ #define DEFAULT_TICK_INTERVAL USToTime(10000) +/* + * When linkerAlwaysPic is true, the runtime linker assume that all object + * files were compiled with -fPIC -fexternal-dynamic-refs and load them + * anywhere in the address space. + */ +#if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS) +#define DEFAULT_LINKER_ALWAYS_PIC true +#else +#define DEFAULT_LINKER_ALWAYS_PIC false +#endif + /* See Note [Synchronization of flags and base APIs] */ typedef struct _MISC_FLAGS { Time tickInterval; /* units: TIME_RESOLUTION */ @@ -197,6 +208,7 @@ typedef struct _MISC_FLAGS { bool generate_stack_trace; bool machineReadable; bool internalCounters; /* See Note [Internal Counter Stats] */ + bool linkerAlwaysPic; /* Assume the object code is always PIC */ StgWord linkerMemBase; /* address to ask the OS for memory * for the linker, NULL ==> off */ } MISC_FLAGS; diff --git a/libraries/base/GHC/RTS/Flags.hsc b/libraries/base/GHC/RTS/Flags.hsc index 12cb828e6a..249bcd5a98 100644 --- a/libraries/base/GHC/RTS/Flags.hsc +++ b/libraries/base/GHC/RTS/Flags.hsc @@ -139,6 +139,7 @@ data MiscFlags = MiscFlags , generateStackTrace :: Bool , machineReadable :: Bool , internalCounters :: Bool + , linkerAlwaysPic :: Bool , linkerMemBase :: Word -- ^ address to ask the OS for memory for the linker, 0 ==> off } deriving ( Show -- ^ @since 4.8.0.0 @@ -444,6 +445,8 @@ getMiscFlags = do (#{peek MISC_FLAGS, machineReadable} ptr :: IO CBool)) <*> (toBool <$> (#{peek MISC_FLAGS, internalCounters} ptr :: IO CBool)) + <*> (toBool <$> + (#{peek MISC_FLAGS, linkerAlwaysPic} ptr :: IO CBool)) <*> #{peek MISC_FLAGS, linkerMemBase} ptr getDebugFlags :: IO DebugFlags diff --git a/rts/Linker.c b/rts/Linker.c index 4c69d0832e..733a609799 100644 --- a/rts/Linker.c +++ b/rts/Linker.c @@ -74,10 +74,6 @@ # include <mach-o/fat.h> #endif -#if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS) -#define ALWAYS_PIC -#endif - #if defined(dragonfly_HOST_OS) #include <sys/tls.h> #endif @@ -203,9 +199,7 @@ static void freeNativeCode_ELF (ObjectCode *nc); * We pick a default address based on the OS, but also make this * configurable via an RTS flag (+RTS -xm) */ -#if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH) - -#if defined(MAP_32BIT) +#if defined(MAP_32BIT) || DEFAULT_LINKER_ALWAYS_PIC // Try to use MAP_32BIT #define MMAP_32BIT_BASE_DEFAULT 0 #else @@ -214,7 +208,6 @@ static void freeNativeCode_ELF (ObjectCode *nc); #endif static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT; -#endif static void ghciRemoveSymbolTable(HashTable *table, const SymbolName* key, ObjectCode *owner) @@ -485,12 +478,10 @@ initLinker_ (int retain_cafs) } # endif -#if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH) if (RtsFlags.MiscFlags.linkerMemBase != 0) { // User-override for mmap_32bit_base mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase; } -#endif if (RTS_LINKER_USE_MMAP) m32_allocator_init(); @@ -1018,29 +1009,30 @@ mmapForLinker (size_t bytes, uint32_t flags, int fd, int offset) void *map_addr = NULL; void *result; size_t size; + uint32_t tryMap32Bit = RtsFlags.MiscFlags.linkerAlwaysPic + ? 0 + : TRY_MAP_32BIT; static uint32_t fixed = 0; IF_DEBUG(linker, debugBelch("mmapForLinker: start\n")); size = roundUpToPage(bytes); -#if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH) mmap_again: if (mmap_32bit_base != 0) { map_addr = mmap_32bit_base; } -#endif IF_DEBUG(linker, debugBelch("mmapForLinker: \tprotection %#0x\n", PROT_EXEC | PROT_READ | PROT_WRITE)); IF_DEBUG(linker, debugBelch("mmapForLinker: \tflags %#0x\n", - MAP_PRIVATE | TRY_MAP_32BIT | fixed | flags)); + MAP_PRIVATE | tryMap32Bit | fixed | flags)); result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE, - MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, offset); + MAP_PRIVATE|tryMap32Bit|fixed|flags, fd, offset); if (result == MAP_FAILED) { sysErrorBelch("mmap %" FMT_Word " bytes at %p",(W_)size,map_addr); @@ -1048,8 +1040,9 @@ mmap_again: return NULL; } -#if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH) - if (mmap_32bit_base != 0) { +#if defined(x86_64_HOST_ARCH) + if (RtsFlags.MiscFlags.linkerAlwaysPic) { + } else if (mmap_32bit_base != 0) { if (result == map_addr) { mmap_32bit_base = (StgWord8*)map_addr + size; } else { @@ -1229,10 +1222,10 @@ void freeObjectCode (ObjectCode *oc) #if defined(NEED_SYMBOL_EXTRAS) && (!defined(x86_64_HOST_ARCH) \ || !defined(mingw32_HOST_OS)) if (RTS_LINKER_USE_MMAP) { - if (!USE_CONTIGUOUS_MMAP && oc->symbol_extras != NULL) { - m32_free(oc->symbol_extras, - sizeof(SymbolExtra) * oc->n_symbol_extras); - } + if (!USE_CONTIGUOUS_MMAP && !RtsFlags.MiscFlags.linkerAlwaysPic && + oc->symbol_extras != NULL) { + m32_free(oc->symbol_extras, sizeof(SymbolExtra) * oc->n_symbol_extras); + } } else { stgFree(oc->symbol_extras); @@ -1533,32 +1526,21 @@ HsInt loadOc (ObjectCode* oc) } /* Note [loadOc orderings] - ocAllocateSymbolsExtras has only two pre-requisites, it must run after - preloadObjectFile and ocVerify. Neither have changed. On most targets - allocating the extras is independent on parsing the section data, so the - order between these two never mattered. + The order of `ocAllocateSymbolExtras` and `ocGetNames` matters. For MachO + and ELF, `ocInit` and `ocGetNames` initialize a bunch of pointers based + on the offset to `oc->image`, but `ocAllocateSymbolExtras` may relocate + the address of `oc->image` and invalidate those pointers. So we must + compute or recompute those pointers after `ocAllocateSymbolExtras`. On Windows, when we have an import library we (for now, as we don't honor the lazy loading semantics of the library and instead GHCi is already lazy) don't use the library after ocGetNames as it just populates the symbol table. Allocating space for jump tables in ocAllocateSymbolExtras would just be a waste then as we'll be stopping further processing of the - library in the next few steps. */ - - /* 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) { - IF_DEBUG(linker, debugBelch("loadOc: ocGetNames_* failed\n")); - return r; - } + library in the next few steps. If necessary, the actual allocation + happens in `ocGetNames_PEi386` and `ocAllocateSymbolExtras_PEi386` simply + set the correct pointers. + */ #if defined(NEED_SYMBOL_EXTRAS) # if defined(OBJFORMAT_MACHO) @@ -1575,7 +1557,24 @@ HsInt loadOc (ObjectCode* oc) debugBelch("loadOc: ocAllocateSymbolExtras_ELF failed\n")); return r; } +# endif + + /* 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) { + IF_DEBUG(linker, debugBelch("loadOc: ocGetNames_* failed\n")); + return r; + } + +# if defined(OBJFORMAT_PEi386) ocAllocateSymbolExtras_PEi386 ( oc ); # endif #endif diff --git a/rts/RtsFlags.c b/rts/RtsFlags.c index ff9635ab24..37eafa5c55 100644 --- a/rts/RtsFlags.c +++ b/rts/RtsFlags.c @@ -236,6 +236,7 @@ void initRtsFlagsDefaults(void) RtsFlags.MiscFlags.generate_dump_file = false; RtsFlags.MiscFlags.machineReadable = false; RtsFlags.MiscFlags.internalCounters = false; + RtsFlags.MiscFlags.linkerAlwaysPic = DEFAULT_LINKER_ALWAYS_PIC; RtsFlags.MiscFlags.linkerMemBase = 0; #if defined(THREADED_RTS) @@ -457,6 +458,11 @@ usage_text[] = { " -e<n> Maximum number of outstanding local sparks (default: 4096)", #endif #if defined(x86_64_HOST_ARCH) +#if !DEFAULT_LINKER_ALWAYS_PIC +" -xp Assume that all object files were compiled with -fPIC", +" -fexternal-dynamic-refs and load them anywhere in the address", +" space", +#endif " -xm Base address to mmap memory in the GHCi linker", " (hex; must be <80000000)", #endif @@ -1502,6 +1508,11 @@ error = true; break; #if defined(x86_64_HOST_ARCH) + case 'p': /* linkerAlwaysPic */ + OPTION_UNSAFE; + RtsFlags.MiscFlags.linkerAlwaysPic = true; + break; + case 'm': /* linkerMemBase */ OPTION_UNSAFE; if (rts_argv[arg][3] != '\0') { diff --git a/rts/linker/Elf.c b/rts/linker/Elf.c index 157e43b385..8954d68965 100644 --- a/rts/linker/Elf.c +++ b/rts/linker/Elf.c @@ -169,6 +169,8 @@ get_shndx_table(Elf_Ehdr* ehdr) void ocInit_ELF(ObjectCode * oc) { + ocDeinit_ELF(oc); + oc->info = (struct ObjectCodeFormatInfo*)stgCallocBytes( 1, sizeof *oc->info, "ocInit_Elf(ObjectCodeFormatInfo)"); @@ -318,6 +320,7 @@ ocDeinit_ELF(ObjectCode * oc) } stgFree(oc->info); + oc->info = NULL; } } @@ -750,7 +753,7 @@ ocGetNames_ELF ( ObjectCode* oc ) start = mem; mapped_start = mem; #else - if (USE_CONTIGUOUS_MMAP) { + if (USE_CONTIGUOUS_MMAP || RtsFlags.MiscFlags.linkerAlwaysPic) { // already mapped. start = oc->image + offset; alloc = SECTION_NOMEM; @@ -1613,9 +1616,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, case COMPAT_R_X86_64_PC32: { -#if defined(ALWAYS_PIC) - barf("R_X86_64_PC32 relocation, but ALWAYS_PIC."); -#else StgInt64 off = value - P; if (off != (Elf64_Sword)off && X86_64_ELF_NONPIC_HACK) { StgInt64 pltAddress = @@ -1632,7 +1632,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, } Elf64_Sword payload = off; memcpy((void*)P, &payload, sizeof(payload)); -#endif break; } @@ -1645,9 +1644,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, case COMPAT_R_X86_64_32: { -#if defined(ALWAYS_PIC) - barf("R_X86_64_32 relocation, but ALWAYS_PIC."); -#else if (value != (Elf64_Word)value && X86_64_ELF_NONPIC_HACK) { StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S) @@ -1663,15 +1659,11 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, } Elf64_Word payload = value; memcpy((void*)P, &payload, sizeof(payload)); -#endif break; } case COMPAT_R_X86_64_32S: { -#if defined(ALWAYS_PIC) - barf("R_X86_64_32S relocation, but ALWAYS_PIC."); -#else if ((StgInt64)value != (Elf64_Sword)value && X86_64_ELF_NONPIC_HACK) { StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S) @@ -1687,7 +1679,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, } Elf64_Sword payload = value; memcpy((void*)P, &payload, sizeof(payload)); -#endif break; } case COMPAT_R_X86_64_REX_GOTPCRELX: @@ -1709,9 +1700,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, #if defined(dragonfly_HOST_OS) case COMPAT_R_X86_64_GOTTPOFF: { -#if defined(ALWAYS_PIC) - barf("R_X86_64_GOTTPOFF relocation, but ALWAYS_PIC."); -#else /* determine the offset of S to the current thread's tls area XXX: Move this to the beginning of function */ @@ -1729,16 +1717,12 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, } Elf64_SWord payload = off; memcpy((void*)P, &payload, sizeof(payload)); -#endif break; } #endif case COMPAT_R_X86_64_PLT32: { -#if defined(ALWAYS_PIC) - barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC."); -#else StgInt64 off = value - P; if (off != (Elf64_Sword)off) { StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S) @@ -1753,7 +1737,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC, } Elf64_Sword payload = off; memcpy((void*)P, &payload, sizeof(payload)); -#endif break; } #endif diff --git a/rts/linker/MachO.c b/rts/linker/MachO.c index 8bdf35ee6b..de7a156403 100644 --- a/rts/linker/MachO.c +++ b/rts/linker/MachO.c @@ -99,6 +99,8 @@ bool ocMprotect_MachO( ObjectCode *oc ); void ocInit_MachO(ObjectCode * oc) { + ocDeinit_MachO(oc); + oc->info = (struct ObjectCodeFormatInfo*)stgCallocBytes( 1, sizeof *oc->info, "ocInit_MachO(ObjectCodeFormatInfo)"); @@ -160,16 +162,19 @@ ocInit_MachO(ObjectCode * oc) void ocDeinit_MachO(ObjectCode * oc) { - if(oc->info->n_macho_symbols > 0) { - stgFree(oc->info->macho_symbols); - } + if (oc->info != NULL) { + if(oc->info->n_macho_symbols > 0) { + stgFree(oc->info->macho_symbols); + } #if defined(aarch64_HOST_ARCH) - freeGot(oc); - for(int i = 0; i < oc->n_sections; i++) { - freeStubs(&oc->sections[i]); - } + freeGot(oc); + for(int i = 0; i < oc->n_sections; i++) { + freeStubs(&oc->sections[i]); + } #endif - stgFree(oc->info); + stgFree(oc->info); + oc->info = NULL; + } } static int diff --git a/rts/linker/SymbolExtras.c b/rts/linker/SymbolExtras.c index 88541f44d0..4c40b10877 100644 --- a/rts/linker/SymbolExtras.c +++ b/rts/linker/SymbolExtras.c @@ -19,6 +19,12 @@ #include "linker/SymbolExtras.h" #include "linker/M32Alloc.h" +#if defined(OBJFORMAT_ELF) +# include "linker/Elf.h" +#elif defined(OBJFORMAT_MACHO) +# include "linker/MachO.h" +#endif + #include <string.h> #if RTS_LINKER_USE_MMAP #include <sys/mman.h> @@ -46,8 +52,24 @@ int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first ) { size_t n; + void* oldImage = oc->image; + + if (count > 0) { + if (!RTS_LINKER_USE_MMAP) { + + // round up to the nearest 4 + int aligned = (oc->fileSize + 3) & ~3; + int misalignment = oc->misalignment; - if (RTS_LINKER_USE_MMAP && USE_CONTIGUOUS_MMAP) { + oc->image -= misalignment; + oc->image = stgReallocBytes( oc->image, + misalignment + + aligned + sizeof (SymbolExtra) * count, + "ocAllocateSymbolExtras" ); + oc->image += misalignment; + + oc->symbol_extras = (SymbolExtra *) (oc->image + aligned); + } else if (USE_CONTIGUOUS_MMAP || RtsFlags.MiscFlags.linkerAlwaysPic) { n = roundUpToPage(oc->fileSize); /* Keep image and symbol_extras contiguous */ @@ -63,42 +85,37 @@ int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first ) oc->imageMapped = true; oc->fileSize = n + (sizeof(SymbolExtra) * count); oc->symbol_extras = (SymbolExtra *) (oc->image + n); - if(mprotect(new, allocated_size, PROT_READ | PROT_EXEC) != 0) { - sysErrorBelch("unable to protect memory"); + if (mprotect(new, allocated_size, + PROT_READ | PROT_WRITE | PROT_EXEC) != 0) { + sysErrorBelch("unable to protect memory"); } } else { oc->symbol_extras = NULL; return 0; } - } - else if( count > 0 ) { - if (RTS_LINKER_USE_MMAP) { - n = roundUpToPage(oc->fileSize); - + } else { oc->symbol_extras = m32_alloc(sizeof(SymbolExtra) * count, 8); if (oc->symbol_extras == NULL) return 0; } - else { - // round up to the nearest 4 - int aligned = (oc->fileSize + 3) & ~3; - int misalignment = oc->misalignment; - - oc->image -= misalignment; - oc->image = stgReallocBytes( oc->image, - misalignment + - aligned + sizeof (SymbolExtra) * count, - "ocAllocateSymbolExtras" ); - oc->image += misalignment; - - oc->symbol_extras = (SymbolExtra *) (oc->image + aligned); - } } if (oc->symbol_extras != NULL) { memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count ); } + // ObjectCodeFormatInfo contains computed addresses based on offset to + // image, if the address of image changes, we need to invalidate + // the ObjectCodeFormatInfo and recompute it. + if (oc->image != oldImage) { +#if defined(OBJFORMAT_MACHO) + ocInit_MachO( oc ); +#endif +#if defined(OBJFORMAT_ELF) + ocInit_ELF( oc ); +#endif + } + oc->first_symbol_extra = first; oc->n_symbol_extras = count; |