Allocate bss section within proper range of other sections

Summary:
This re-applies {D5195} and {D5235}, they were reverted as part of diff
stack to unbreak i386. The proper fix is done in {D5289}.

Allocate bss section within proper range of other sections:

* when `+RTS -xp` is passed, allocate it contiguously as we did for
  jump islands
* when we mmap the code to lower 2Gb, we should allocate bss section
  there too

Test Plan:
1. `./validate`

2.

with

```
DYNAMIC_GHC_PROGRAMS = NO
DYNAMIC_BY_DEFAULT = NO
```

`TEST="T15729" make test` passed in both linux (both i386 and x86_64) and macos.

3.

Also test in a use case where we used to encouter error like:

```
ghc-iserv-prof: R_X86_64_PC32 relocation out of range: (noname) =
b90282ba
```

and now, everything works fine.

Reviewers: simonmar, bgamari, angerman, erikd

Reviewed By: simonmar

Subscribers: rwbarton, carter

GHC Trac Issues: #15729

Differential Revision: https://phabricator.haskell.org/D5290

8 files changed, 73 insertions, 60 deletions

diff --git a/rts/linker/Elf.c b/rts/linker/Elf.c
index ae85d73170..6de531a920 100644
--- a/rts/linker/Elf.c
+++ b/rts/linker/Elf.c
@@ -579,7 +579,7 @@ ocVerifyImage_ELF ( ObjectCode* oc )
 /* Figure out what kind of section it is.  Logic derived from
    Figure 1.14 ("Special Sections") of the ELF document
    ("Portable Formats Specification, Version 1.1"). */
-static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
+static SectionKind getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
 {
     *is_bss = false;
 
@@ -681,23 +681,31 @@ ocGetNames_ELF ( ObjectCode* oc )
       StgWord mapped_size = 0, mapped_offset = 0;
       StgWord size = shdr[i].sh_size;
       StgWord offset = shdr[i].sh_offset;
+      StgWord align = shdr[i].sh_addralign;
 
       if (is_bss && size > 0) {
          /* This is a non-empty .bss section.  Allocate zeroed space for
             it, and set its .sh_offset field such that
             ehdrC + .sh_offset == addr_of_zeroed_space.  */
-#if defined(NEED_GOT)
-          /* always use mmap if we use GOT slots.  Otherwise the malloced
-           * address might be out of range for sections that are mmaped.
-           */
-          alloc = SECTION_MMAP;
-          start = mmap(NULL, size,
-                       PROT_READ | PROT_WRITE | PROT_EXEC,
-                       MAP_ANON | MAP_PRIVATE,
-                       -1, 0);
-          mapped_start = start;
-          mapped_offset = 0;
-          mapped_size = roundUpToPage(size);
+#if defined(NEED_GOT) || RTS_LINKER_USE_MMAP
+          if (USE_CONTIGUOUS_MMAP || RtsFlags.MiscFlags.linkerAlwaysPic) {
+              /* The space for bss sections is already preallocated */
+              ASSERT(oc->bssBegin != NULL);
+              alloc = SECTION_NOMEM;
+              start =
+                oc->image + roundUpToAlign(oc->bssBegin - oc->image, align);
+              oc->bssBegin = (char*)start + size;
+              ASSERT(oc->bssBegin <= oc->bssEnd);
+          } else {
+              /* Use mmapForLinker to allocate .bss, otherwise the malloced
+               * address might be out of range for sections that are mmaped.
+               */
+              alloc = SECTION_MMAP;
+              start = mmapForLinker(size, MAP_ANONYMOUS, -1, 0);
+              mapped_start = start;
+              mapped_offset = 0;
+              mapped_size = roundUpToPage(size);
+          }
 #else
           alloc = SECTION_MALLOC;
           start = stgCallocBytes(1, size, "ocGetNames_ELF(BSS)");
@@ -1874,22 +1882,27 @@ int ocRunInit_ELF( ObjectCode *oc )
 
 #if defined(NEED_SYMBOL_EXTRAS)
 
-int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
+int ocAllocateExtras_ELF( ObjectCode *oc )
 {
-  Elf_Ehdr *ehdr;
-  Elf_Shdr* shdr;
-  Elf_Word i, shnum;
-
-  ehdr = (Elf_Ehdr *) oc->image;
-  shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
-
-  shnum = elf_shnum(ehdr);
-
-  for( i = 0; i < shnum; i++ )
-    if( shdr[i].sh_type == SHT_SYMTAB )
-      break;
+  Elf_Ehdr *ehdr = (Elf_Ehdr *) oc->image;
+  Elf_Shdr* shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
+  Elf_Shdr* symtab = NULL;
+  Elf_Word shnum = elf_shnum(ehdr);
+  int bssSize = 0;
+
+  for (Elf_Word i = 0; i < shnum; ++i) {
+    if(shdr[i].sh_type == SHT_SYMTAB) {
+      symtab = &shdr[i];
+    } else {
+      int isBss = 0;
+      getSectionKind_ELF(&shdr[i], &isBss);
+      if (isBss && shdr[i].sh_size > 0) {
+        bssSize += roundUpToAlign(shdr[i].sh_size, shdr[i].sh_addralign);
+      }
+    }
+  }
 
-  if( i == shnum )
+  if (symtab == NULL)
   {
     // Not having a symbol table is not in principle a problem.
     // When an object file has no symbols then the 'strip' program
@@ -1899,15 +1912,15 @@ int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
     return 1;
   }
 
-  if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
+  if( symtab->sh_entsize != sizeof( Elf_Sym ) )
   {
     errorBelch( "The entry size (%d) of the symtab isn't %d\n",
-      (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
+      (int) symtab->sh_entsize, (int) sizeof( Elf_Sym ) );
 
     return 0;
   }
 
-  return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
+  return ocAllocateExtras(oc, symtab->sh_size / sizeof( Elf_Sym ), 0, bssSize);
 }
 
 #endif /* NEED_SYMBOL_EXTRAS */
diff --git a/rts/linker/Elf.h b/rts/linker/Elf.h
index b0d6638e6a..30c993b98c 100644
--- a/rts/linker/Elf.h
+++ b/rts/linker/Elf.h
@@ -13,6 +13,6 @@ int ocVerifyImage_ELF    ( ObjectCode* oc );
 int ocGetNames_ELF       ( ObjectCode* oc );
 int ocResolve_ELF        ( ObjectCode* oc );
 int ocRunInit_ELF        ( ObjectCode* oc );
-int ocAllocateSymbolExtras_ELF( ObjectCode *oc );
+int ocAllocateExtras_ELF ( ObjectCode *oc );
 
 #include "EndPrivate.h"
diff --git a/rts/linker/MachO.c b/rts/linker/MachO.c
index 4a1204dde9..ca5befca77 100644
--- a/rts/linker/MachO.c
+++ b/rts/linker/MachO.c
@@ -187,19 +187,22 @@ resolveImports(
 #if defined(x86_64_HOST_ARCH) || defined(aarch64_HOST_ARCH)
 
 int
-ocAllocateSymbolExtras_MachO(ObjectCode* oc)
+ocAllocateExtras_MachO(ObjectCode* oc)
 {
-    IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO: start\n"));
+    IF_DEBUG(linker, debugBelch("ocAllocateExtras_MachO: start\n"));
 
     if (NULL != oc->info->symCmd) {
-        IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO: allocate %d symbols\n", oc->info->symCmd->nsyms));
-        IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO: done\n"));
-        return ocAllocateSymbolExtras(oc, oc->info->symCmd->nsyms, 0);
+        IF_DEBUG(linker,
+            debugBelch("ocAllocateExtras_MachO: allocate %d symbols\n",
+                oc->info->symCmd->nsyms));
+        IF_DEBUG(linker, debugBelch("ocAllocateExtras_MachO: done\n"));
+        return ocAllocateExtras(oc, oc->info->symCmd->nsyms, 0, 0);
     }
 
-    IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO: allocated no symbols\n"));
-    IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO: done\n"));
-    return ocAllocateSymbolExtras(oc,0,0);
+    IF_DEBUG(linker,
+        debugBelch("ocAllocateExtras_MachO: allocated no symbols\n"));
+    IF_DEBUG(linker, debugBelch("ocAllocateExtras_MachO: done\n"));
+    return ocAllocateExtras(oc, 0, 0, 0);
 }
 
 #else
diff --git a/rts/linker/MachO.h b/rts/linker/MachO.h
index b495c2b9b1..4fb58e8668 100644
--- a/rts/linker/MachO.h
+++ b/rts/linker/MachO.h
@@ -13,7 +13,7 @@ int ocGetNames_MachO       ( ObjectCode* oc );
 int ocResolve_MachO        ( ObjectCode* oc );
 int ocRunInit_MachO        ( ObjectCode* oc );
 int machoGetMisalignment( FILE * );
-int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
+int ocAllocateExtras_MachO ( ObjectCode* oc );
 
 #if defined(powerpc_HOST_ARCH)
 void machoInitSymbolsWithoutUnderscore( void );
diff --git a/rts/linker/PEi386.c b/rts/linker/PEi386.c
index cc92fa78a2..6cf0d52d39 100644
--- a/rts/linker/PEi386.c
+++ b/rts/linker/PEi386.c
@@ -1778,7 +1778,7 @@ ocGetNames_PEi386 ( ObjectCode* oc )
  * so simply set correct pointer here.
  */
 bool
-ocAllocateSymbolExtras_PEi386 ( ObjectCode* oc )
+ocAllocateExtras_PEi386 ( ObjectCode* oc )
 {
    /* If the ObjectCode was unloaded we don't need a trampoline, it's likely
       an import library so we're discarding it earlier.  */
diff --git a/rts/linker/PEi386.h b/rts/linker/PEi386.h
index eb5bec8b78..538f132ab5 100644
--- a/rts/linker/PEi386.h
+++ b/rts/linker/PEi386.h
@@ -57,7 +57,7 @@ bool ocRunInit_PEi386     ( ObjectCode *oc );
 bool ocGetNames_PEi386    ( ObjectCode* oc );
 bool ocVerifyImage_PEi386 ( ObjectCode* oc );
 SymbolAddr *lookupSymbol_PEi386(SymbolName *lbl);
-bool ocAllocateSymbolExtras_PEi386 ( ObjectCode* oc );
+bool ocAllocateExtras_PEi386 ( ObjectCode* oc );
 SymbolAddr *lookupSymbolInDLLs ( const SymbolName* lbl );
 /* See Note [mingw-w64 name decoration scheme] */
 /* We use myindex to calculate array addresses, rather than
diff --git a/rts/linker/SymbolExtras.c b/rts/linker/SymbolExtras.c
index 4c40b10877..a9e4c37967 100644
--- a/rts/linker/SymbolExtras.c
+++ b/rts/linker/SymbolExtras.c
@@ -31,10 +31,11 @@
 #endif /* RTS_LINKER_USE_MMAP */
 
 /*
-  ocAllocateSymbolExtras
+  ocAllocateExtras
 
   Allocate additional space at the end of the object file image to make room
-  for jump islands (powerpc, x86_64, arm) and GOT entries (x86_64).
+  for jump islands (powerpc, x86_64, arm), GOT entries (x86_64) and
+  bss sections.
 
   PowerPC relative branch instructions have a 24 bit displacement field.
   As PPC code is always 4-byte-aligned, this yields a +-32MB range.
@@ -49,12 +50,11 @@
   filled in by makeSymbolExtra below.
 */
 
-int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
+int ocAllocateExtras(ObjectCode* oc, int count, int first, int bssSize)
 {
-  size_t n;
   void* oldImage = oc->image;
 
-  if (count > 0) {
+  if (count > 0 || bssSize > 0) {
     if (!RTS_LINKER_USE_MMAP) {
 
       // round up to the nearest 4
@@ -65,16 +65,15 @@ int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
       oc->image = stgReallocBytes( oc->image,
                                misalignment +
                                aligned + sizeof (SymbolExtra) * count,
-                               "ocAllocateSymbolExtras" );
+                               "ocAllocateExtras" );
       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 */
-
-      size_t allocated_size = n + (sizeof(SymbolExtra) * count);
+      /* Keep image, bssExtras and symbol_extras contiguous */
+      size_t n = roundUpToPage(oc->fileSize);
+      bssSize = roundUpToAlign(bssSize, 8);
+      size_t allocated_size = n + bssSize + (sizeof(SymbolExtra) * count);
       void *new = mmapForLinker(allocated_size, MAP_ANONYMOUS, -1, 0);
       if (new) {
           memcpy(new, oc->image, oc->fileSize);
@@ -83,12 +82,10 @@ int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
           }
           oc->image = new;
           oc->imageMapped = true;
-          oc->fileSize = n + (sizeof(SymbolExtra) * count);
-          oc->symbol_extras = (SymbolExtra *) (oc->image + n);
-          if (mprotect(new, allocated_size,
-                       PROT_READ | PROT_WRITE | PROT_EXEC) != 0) {
-            sysErrorBelch("unable to protect memory");
-          }
+          oc->fileSize = allocated_size;
+          oc->symbol_extras = (SymbolExtra *) (oc->image + n + bssSize);
+          oc->bssBegin = oc->image + n;
+          oc->bssEnd = oc->image + n + bssSize;
       }
       else {
           oc->symbol_extras = NULL;
diff --git a/rts/linker/SymbolExtras.h b/rts/linker/SymbolExtras.h
index 4974c06e7d..af828e66fa 100644
--- a/rts/linker/SymbolExtras.h
+++ b/rts/linker/SymbolExtras.h
@@ -7,7 +7,7 @@
 
 #if defined(NEED_SYMBOL_EXTRAS)
 
-int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first );
+int ocAllocateExtras(ObjectCode* oc, int count, int first, int bssSize);
 
 #if defined(arm_HOST_ARCH)
 SymbolExtra* makeArmSymbolExtra( ObjectCode const* oc,