RISC-V: adjust logic to avoid register name symbols

Special casing GPR names in my_getSmallExpression() leads to a number of
inconsistencies. Generalize this by utilizing the md_parse_name() hook,
limited to when instruction operands are being parsed (really: probed).
Then both the GPR lookup there and the yet more ad hoc workaround for
PR/gas 29940 can be removed (including its extension needed for making
the compressed form JAL work again).

2 files changed, 98 insertions, 27 deletions

diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
index c99bab9..a94d4f6 100644
--- a/gas/config/tc-riscv.c
+++ b/gas/config/tc-riscv.c
@@ -171,6 +171,8 @@ static enum float_abi float_abi = FLOAT_ABI_DEFAULT;
 
 static unsigned elf_flags = 0;
 
+static bool probing_insn_operands;
+
 /* Set the default_isa_spec.  Return 0 if the spec isn't supported.
    Otherwise, return 1.  */
 
@@ -2242,21 +2244,10 @@ my_getSmallExpression (expressionS *ep, bfd_reloc_code_real_type *reloc,
 		       char *str, const struct percent_op_match *percent_op)
 {
   size_t reloc_index;
-  unsigned crux_depth, str_depth, regno;
+  unsigned crux_depth, str_depth;
+  bool orig_probing = probing_insn_operands;
   char *crux;
 
-  /* First, check for integer registers.  No callers can accept a reg, but
-     we need to avoid accidentally creating a useless undefined symbol below,
-     if this is an instruction pattern that can't match.  A glibc build fails
-     if this is removed.  */
-  if (reg_lookup (&str, RCLASS_GPR, &regno))
-    {
-      ep->X_op = O_register;
-      ep->X_add_number = regno;
-      expr_parse_end = str;
-      return 0;
-    }
-
   /* Search for the start of the main expression.
 
      End the loop with CRUX pointing to the start of the main expression and
@@ -2288,9 +2279,17 @@ my_getSmallExpression (expressionS *ep, bfd_reloc_code_real_type *reloc,
        return 0;
     }
 
+  /* Anything inside parentheses or subject to a relocation operator cannot
+     be a register and hence can be treated the same as operands to
+     directives (other than .insn).  */
+  if (str_depth || reloc_index)
+    probing_insn_operands = false;
+
   my_getExpression (ep, crux);
   str = expr_parse_end;
 
+  probing_insn_operands = orig_probing;
+
   /* Match every open bracket.  */
   while (crux_depth > 0 && (*str == ')' || *str == ' ' || *str == '\t'))
     if (*str++ == ')')
@@ -2476,6 +2475,13 @@ riscv_is_priv_insn (insn_t insn)
 	  || ((insn ^ MATCH_SFENCE_VM) & MASK_SFENCE_VM) == 0);
 }
 
+static symbolS *deferred_sym_rootP;
+static symbolS *deferred_sym_lastP;
+/* Since symbols can't easily be freed, try to recycle ones which weren't
+   committed.  */
+static symbolS *orphan_sym_rootP;
+static symbolS *orphan_sym_lastP;
+
 /* This routine assembles an instruction into its binary format.  As a
    side effect, it sets the global variable imm_reloc to the type of
    relocation to do if one of the operands is an address expression.  */
@@ -2511,6 +2517,8 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
 
   insn = (struct riscv_opcode *) str_hash_find (hash, str);
 
+  probing_insn_operands = true;
+
   asargStart = asarg;
   for ( ; insn && insn->name && strcmp (insn->name, str) == 0; insn++)
     {
@@ -2527,6 +2535,17 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
       /* Reset error message of the previous round.  */
       error.msg = _("illegal operands");
       error.missing_ext = NULL;
+
+      /* Purge deferred symbols from the previous round, if any.  */
+      while (deferred_sym_rootP)
+	{
+	  symbolS *sym = deferred_sym_rootP;
+
+	  symbol_remove (sym, &deferred_sym_rootP, &deferred_sym_lastP);
+	  symbol_append (sym, orphan_sym_lastP, &orphan_sym_rootP,
+			 &orphan_sym_lastP);
+	}
+
       create_insn (ip, insn);
 
       imm_expr->X_op = O_absent;
@@ -2581,9 +2600,22 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
 		}
 	      if (*asarg != '\0')
 		break;
+
 	      /* Successful assembly.  */
 	      error.msg = NULL;
 	      insn_with_csr = false;
+
+	      /* Commit deferred symbols, if any.  */
+	      while (deferred_sym_rootP)
+		{
+		  symbolS *sym = deferred_sym_rootP;
+
+		  symbol_remove (sym, &deferred_sym_rootP,
+				 &deferred_sym_lastP);
+		  symbol_append (sym, symbol_lastP, &symbol_rootP,
+				 &symbol_lastP);
+		  symbol_table_insert (sym);
+		}
 	      goto out;
 
 	    case 'C': /* RVC */
@@ -2787,8 +2819,6 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
 		case 'p':
 		  goto branch;
 		case 'a':
-		  if (oparg == insn->args + 1)
-		    goto jump_check_gpr;
 		  goto jump;
 		case 'S': /* Floating-point RS1 x8-x15.  */
 		  if (!reg_lookup (&asarg, RCLASS_FPR, &regno)
@@ -3292,18 +3322,6 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
 	      continue;
 
 	    case 'a': /* 20-bit PC-relative offset.  */
-	      /* Like in my_getSmallExpression() we need to avoid emitting
-		 a stray undefined symbol if the 1st JAL entry doesn't match,
-		 but the 2nd (with 2 operands) might.  */
-	      if (oparg == insn->args)
-		{
-	    jump_check_gpr:
-		  asargStart = asarg;
-		  if (reg_lookup (&asarg, RCLASS_GPR, NULL)
-		      && (*asarg == ',' || (ISSPACE (*asarg) && asarg[1] == ',')))
-		    break;
-		  asarg = asargStart;
-		}
 	    jump:
 	      my_getExpression (imm_expr, asarg);
 	      asarg = expr_parse_end;
@@ -3542,6 +3560,8 @@ riscv_ip (char *str, struct riscv_cl_insn *ip, expressionS *imm_expr,
   if (save_c)
     *(asargStart  - 1) = save_c;
 
+  probing_insn_operands = false;
+
   return error;
 }
 
@@ -3838,6 +3858,53 @@ riscv_after_parse_args (void)
     flag_dwarf_cie_version = 3;
 }
 
+bool riscv_parse_name (const char *name, struct expressionS *ep,
+		       enum expr_mode mode)
+{
+  unsigned int regno;
+  symbolS *sym;
+
+  if (!probing_insn_operands)
+    return false;
+
+  gas_assert (mode == expr_normal);
+
+  regno = reg_lookup_internal (name, RCLASS_GPR);
+  if (regno == (unsigned int)-1)
+    return false;
+
+  if (symbol_find (name) != NULL)
+    return false;
+
+  /* Create a symbol without adding it to the symbol table yet.
+     Insertion will happen only once we commit to using the insn
+     we're probing operands for.  */
+  for (sym = deferred_sym_rootP; sym; sym = symbol_next (sym))
+    if (strcmp (name, S_GET_NAME (sym)) == 0)
+      break;
+  if (!sym)
+    {
+      for (sym = orphan_sym_rootP; sym; sym = symbol_next (sym))
+	if (strcmp (name, S_GET_NAME (sym)) == 0)
+	  {
+	    symbol_remove (sym, &orphan_sym_rootP, &orphan_sym_lastP);
+	    break;
+	  }
+      if (!sym)
+	sym = symbol_create (name, undefined_section,
+			     &zero_address_frag, 0);
+
+      symbol_append (sym, deferred_sym_lastP, &deferred_sym_rootP,
+		     &deferred_sym_lastP);
+    }
+
+  ep->X_op = O_symbol;
+  ep->X_add_symbol = sym;
+  ep->X_add_number = 0;
+
+  return true;
+}
+
 long
 md_pcrel_from (fixS *fixP)
 {
diff --git a/gas/config/tc-riscv.h b/gas/config/tc-riscv.h
index 111492a..0c70c7d 100644
--- a/gas/config/tc-riscv.h
+++ b/gas/config/tc-riscv.h
@@ -123,6 +123,10 @@ extern void riscv_elf_final_processing (void);
 /* Adjust debug_line after relaxation.  */
 #define DWARF2_USE_FIXED_ADVANCE_PC 1
 
+#define md_parse_name(name, exp, mode, c) \
+  riscv_parse_name (name, exp, mode)
+bool riscv_parse_name (const char *, struct expressionS *, enum expr_mode);
+
 #define md_finish riscv_md_finish
 #define CONVERT_SYMBOLIC_ATTRIBUTE riscv_convert_symbolic_attribute