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

Summary:
This re-applies {D5195} with fixes for i386:
* Fix unused label warnings, see {D5230} or {D5273}
* Fix a silly bug introduced by moving `#if`

{P190}

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:
See {D5195}, also test under i386:

```
$ uname -a
Linux watashi-arch32 4.18.5-arch1-1.0-ARCH #1 SMP PREEMPT Tue Aug 28
20:45:30 CEST 2018 i686 GNU/Linux
$ cd testsuite/tests/th/ && make test
...
```

will run `./validate` on stacked diff.

Reviewers: simonmar, bgamari, alpmestan, trommler, hvr, erikd

Reviewed By: simonmar

Subscribers: rwbarton, carter

Differential Revision: https://phabricator.haskell.org/D5289

1 files changed, 43 insertions, 41 deletions

diff --git a/rts/Linker.c b/rts/Linker.c
index ac030af837..296b6441ea 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 +1000,32 @@ 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)
+#if defined(x86_64_HOST_ARCH)
 mmap_again:
+#endif
 
    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 +1033,9 @@ 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 +1203,10 @@ 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 +1499,21 @@ 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 +1530,26 @@ HsInt loadOc (ObjectCode* oc)
                 debugBelch("loadOc: ocAllocateSymbolExtras_ELF failed\n"));
        return r;
    }
+#  endif
+#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(NEED_SYMBOL_EXTRAS)
+#  if defined(OBJFORMAT_PEi386)
    ocAllocateSymbolExtras_PEi386 ( oc );
 #  endif
 #endif
@@ -1833,4 +1836,3 @@ addSection (Section *s, SectionKind kind, SectionAlloc alloc,
                        start, (void*)((StgWord)start + size),
                        size, kind ));
 }
-