Use 32-bits pointers to C code

On a 64-bit machine, we only need 32 bits to store a pointer to
the C code that implements a BEAM instruction. Refactor the code
to only use the lower 32 bits of each instruction word, and take
care to preserve the high 32 bits.

3 files changed, 44 insertions, 16 deletions

diff --git a/erts/emulator/beam/beam_bp.c b/erts/emulator/beam/beam_bp.c
index c81380c14d..871670e8c3 100644
--- a/erts/emulator/beam/beam_bp.c
+++ b/erts/emulator/beam/beam_bp.c
@@ -419,9 +419,11 @@ erts_install_breakpoints(BpFunctions* f)
 
     for (i = 0; i < n; i++) {
 	ErtsCodeInfo* ci = f->matching[i].ci;
-        BeamInstr *pc = erts_codeinfo_to_code(ci);
 	GenericBp* g = ci->u.gen_bp;
-	if (*pc != br && g) {
+        BeamInstr volatile *pc = erts_codeinfo_to_code(ci);
+        BeamInstr instr = *pc;
+
+	if (!BeamIsOpCode(instr, op_i_generic_breakpoint) && g) {
 	    Module* modp = f->matching[i].mod;
 
 	    /*
@@ -435,11 +437,16 @@ erts_install_breakpoints(BpFunctions* f)
 	    /*
 	     * The following write is not protected by any lock. We
 	     * assume that the hardware guarantees that a write of an
-	     * aligned word-size (or half-word) writes is atomic
-	     * (i.e. that other processes executing this code will not
-	     * see a half pointer).
+	     * aligned word-size writes is atomic (i.e. that other
+	     * processes executing this code will not see a half
+	     * pointer).
+             *
+             * The contents of *pc is marked 'volatile' to ensure that
+             * the compiler will do a single full-word write, and not
+             * try any fancy optimizations to write a half word.
 	     */
-	    *pc = br;
+            instr = BeamSetCodeAddr(instr, br);
+            *pc = instr;
 	    modp->curr.num_breakpoints++;
 	}
     }
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 9a77c63390..6a45087a34 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -50,14 +50,14 @@
 #if defined(NO_JUMP_TABLE)
 #  define OpCase(OpCode)    case op_##OpCode
 #  define CountCase(OpCode) case op_count_##OpCode
-#  define IsOpCode(InstrWord, OpCode)  ((InstrWord) == (BeamInstr)op_##OpCode)
-#  define Goto(Rel) {Go = (Rel); goto emulator_loop;}
+#  define IsOpCode(InstrWord, OpCode)  (BeamCodeAddr(InstrWord) == (BeamInstr)op_##OpCode)
+#  define Goto(Rel)         {Go = BeamCodeAddr(Rel); goto emulator_loop;}
 #else
 #  define OpCase(OpCode)    lb_##OpCode
 #  define CountCase(OpCode) lb_count_##OpCode
-#  define IsOpCode(InstrWord, OpCode)  ((InstrWord) == (BeamInstr)&&lb_##OpCode)
-#  define Goto(Rel) goto *((void *)Rel)
-#  define LabelAddr(Label) &&Label
+#  define IsOpCode(InstrWord, OpCode)  (BeamCodeAddr(InstrWord) == (BeamInstr)&&lb_##OpCode)
+#  define Goto(Rel)         goto *((void *)BeamCodeAddr(Rel))
+#  define LabelAddr(Label)  &&Label
 #endif
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -131,11 +131,11 @@ do {                                     \
 
 /* We don't check the range if an ordinary switch is used */
 #ifdef NO_JUMP_TABLE
-#define VALID_INSTR(IP) ((UWord)(IP) < (NUMBER_OF_OPCODES*2+10))
+#  define VALID_INSTR(IP) (BeamCodeAddr(IP) < (NUMBER_OF_OPCODES*2+10))
 #else
-#define VALID_INSTR(IP) \
-   ((SWord)LabelAddr(emulator_loop) <= (SWord)(IP) && \
-    (SWord)(IP) < (SWord)LabelAddr(end_emulator_loop))
+#  define VALID_INSTR(IP) \
+    ((BeamInstr)LabelAddr(emulator_loop) <= BeamCodeAddr(IP) && \
+     BeamCodeAddr(IP) < (BeamInstr)LabelAddr(end_emulator_loop))
 #endif /* NO_JUMP_TABLE */
 
 #define SET_CP(p, ip)           \
@@ -1006,6 +1006,18 @@ init_emulator_finish(void)
      int i;
      Export* ep;
 
+#if defined(ARCH_64) && defined(CODE_MODEL_SMALL)
+     for (i = 0; i < NUMBER_OF_OPCODES; i++) {
+         BeamInstr instr = BeamOpCodeAddr(i);
+         if (instr >= (1ull << 32)) {
+             erts_exit(ERTS_ERROR_EXIT,
+                       "This run-time was supposed be compiled with all code below 2Gb,\n"
+                       "but the instruction '%s' is located at %016lx.\n",
+                       opc[i].name, instr);
+         }
+     }
+#endif
+
      beam_apply[0]             = BeamOpCodeAddr(op_i_apply);
      beam_apply[1]             = BeamOpCodeAddr(op_normal_exit);
      beam_exit[0]              = BeamOpCodeAddr(op_error_action_code);
diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h
index 661538eadd..76980b5871 100644
--- a/erts/emulator/beam/erl_vm.h
+++ b/erts/emulator/beam/erl_vm.h
@@ -209,6 +209,15 @@ extern void** beam_ops;
 #  define BeamOpCodeAddr(OpCode) ((BeamInstr)beam_ops[(OpCode)])
 #endif
 
-#define BeamIsOpCode(InstrWord, OpCode) ((InstrWord) == BeamOpCodeAddr(OpCode))
+#if defined(ARCH_64) && defined(CODE_MODEL_SMALL)
+#  define BeamCodeAddr(InstrWord) ((BeamInstr)(Uint32)(InstrWord))
+#  define BeamSetCodeAddr(InstrWord, Addr) (((InstrWord) & ~((1ull << 32)-1)) | (Addr))
+#  define BeamExtraData(InstrWord) ((InstrWord) >> 32)
+#else
+#  define BeamCodeAddr(InstrWord) ((BeamInstr)(InstrWord))
+#  define BeamSetCodeAddr(InstrWord, Addr) (Addr)
+#endif
+
+#define BeamIsOpCode(InstrWord, OpCode) (BeamCodeAddr(InstrWord) == BeamOpCodeAddr(OpCode))
 
 #endif	/* __ERL_VM_H__ */