Add a RTS option -xp to load PIC object anywhere in address space

Add a RTS option -xp to load PIC object anywhere in address space. We do
this by relaxing the requirement of <0x80000000 result of
`mmapForLinker` and implying USE_CONTIGUOUS_MMAP.

We also need to change calls to `ocInit` and `ocGetNames` to avoid
dangling pointers when the address of `oc->image` is changed by
`ocAllocateSymbolExtra`.

Test Plan:
```
$ uname -a
Linux localhost 4.18.8-arch1-1-ARCH #1 SMP PREEMPT Sat Sep 15 20:34:48
UTC 2018 x86_64 GNU/Linux
$ cat mk/build.mk
DYNAMIC_GHC_PROGRAMS = NO
DYNAMIC_BY_DEFAULT = NO

GhcRTSWays += thr_debug
EXTRA_HC_OPTS += -debug
WAY_p_HC_OPTS += -fPIC -fexternal-dynamic-refs
$ inplace/bin/ghc-stage2 --interactive -prof +RTS -xp
GHCi, version 8.7.20180928: http://www.haskell.org/ghc/  :? for help
ghc-stage2: R_X86_64_32 relocation out of range:
ghczmprim_GHCziTypes_ZMZN_closure = 7f690bffab59
Recompile
/data/users/watashi/ghc/libraries/ghc-prim/dist-install/build/HSghc-prim
-0.5.3.o with -fPIC -fexternal-dynamic-refs.
ghc-stage2: unable to load package `ghc-prim-0.5.3'
$ strace -f -e open,mmap inplace/bin/ghc-stage2 --interactive -prof
-fexternal-interpreter  -opti+RTS -opti-xp
...
[pid 1355283]
open("/data/users/watashi/ghc/libraries/base/dist-install/build/libHSbas
e-4.12.0.0_p.a", O_RDONLY) = 14
[pid 1355283] mmap(NULL, 8192, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6a84842000
[pid 1355283]
open("/data/users/watashi/ghc/libraries/base/dist-install/build/libHSbas
e-4.12.0.0_p.a", O_RDONLY) = 14
[pid 1355283] mmap(NULL, 8192, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6a84676000
...
Prelude> System.Posix.Process.getProcessID
...
[pid 1355283]
open("/data/users/watashi/ghc/libraries/unix/dist-install/build/libHSuni
x-2.7.2.2_p.a", O_RDONLY) = 14
[pid 1355283] mmap(NULL, 45056, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6a67d60000
[pid 1355283]
open("/data/users/watashi/ghc/libraries/unix/dist-install/build/libHSuni
x-2.7.2.2_p.a", O_RDONLY) = 14
[pid 1355283] mmap(NULL, 57344, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f6a67d52000
...
```

```
$ uname -a
Darwin watashis-iMac.local 18.0.0 Darwin Kernel Version 18.0.0: Wed Aug
22 20:13:40 PDT 2018; root:xnu-4903.201.2~1/RELEASE_X86_64 x86_64
$ mv
/Users/watashi/gao/ghc/libraries/integer-gmp/dist-install/build/HSintege
r-gmp-1.0.2.0.o{,._DISABLE_GHC_ISSUE_15105}
$ inplace/bin/ghc-stage2 --interactive  +RTS -xp
GHCi, version 8.7.20181003: http://www.haskell.org/ghc/  :? for help
Prelude> System.Posix.Process.getProcessID
42791
Prelude> Data.Set.fromList [1 .. 10]
fromList [1,2,3,4,5,6,7,8,9,10]
Prelude>
Leaving GHCi.
$ inplace/bin/ghc-stage2 --interactive -prof -fexternal-interpreter
GHCi, version 8.7.20181003: http://www.haskell.org/ghc/  :? for help
Prelude> System.Posix.Process.getProcessID
42806
Prelude> Data.Set.fromList [1 .. 10]
fromList [1,2,3,4,5,6,7,8,9,10]
Prelude>
Leaving GHCi.
```

Also test with something that used to hit the 2Gb limit and it loads
and runs without problem.

Reviewers: simonmar, bgamari, angerman, Phyx, hvr, erikd

Reviewed By: simonmar

Subscribers: rwbarton, carter

Differential Revision: https://phabricator.haskell.org/D5195

5 files changed, 106 insertions, 92 deletions

diff --git a/rts/Linker.c b/rts/Linker.c
index 826269738b..b42a0de9c3 100644
--- a/rts/Linker.c
+++ b/rts/Linker.c
@@ -72,10 +72,6 @@
 #  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
@@ -212,9 +208,7 @@ int ocTryLoad( ObjectCode* oc );
  * 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
@@ -223,7 +217,6 @@ int ocTryLoad( ObjectCode* oc );
 #endif
 
 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
-#endif
 
 static void ghciRemoveSymbolTable(HashTable *table, const SymbolName* key,
     ObjectCode *owner)
@@ -496,12 +489,10 @@ 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();
@@ -1009,29 +1000,30 @@ 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);
@@ -1039,8 +1031,9 @@ 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 {
@@ -1208,10 +1201,10 @@ 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);
@@ -1504,32 +1497,21 @@ 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)
@@ -1546,7 +1528,24 @@ 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
@@ -1830,4 +1829,3 @@ addSection (Section *s, SectionKind kind, SectionAlloc alloc,
                        start, (void*)((StgWord)start + size),
                        size, kind ));
 }
-
diff --git a/rts/RtsFlags.c b/rts/RtsFlags.c
index 6a72e67859..6aecd90ca4 100644
--- a/rts/RtsFlags.c
+++ b/rts/RtsFlags.c
@@ -235,6 +235,7 @@ 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)
@@ -453,6 +454,11 @@ 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
@@ -1475,6 +1481,11 @@ 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 fd24a92630..fe87aed0b0 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;
     }
 }
 
@@ -754,7 +757,7 @@ 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;
@@ -1584,9 +1587,6 @@ 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 =
@@ -1603,7 +1603,6 @@ do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
           }
           Elf64_Sword payload = off;
           memcpy((void*)P, &payload, sizeof(payload));
-#endif
           break;
       }
 
@@ -1616,9 +1615,6 @@ 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)
@@ -1634,15 +1630,11 @@ 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)
@@ -1658,7 +1650,6 @@ 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:
@@ -1680,9 +1671,6 @@ 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 */
@@ -1700,16 +1688,12 @@ 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)
@@ -1724,7 +1708,6 @@ 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 5812e89cda..ff8ef7a1e4 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;