ld script expression parsing

Message ID 20210201110456.GI26219@bubble.grove.modra.org
State New
Headers show
Series
  • ld script expression parsing
Related show

Commit Message

Alan Modra via Binutils Feb. 1, 2021, 11:04 a.m.
Parsing symbol or file/section names in ld linker scripts is a little
complicated.  Inside SECTIONS, a name might be the start of an
expression or an output section.  Is ".foo=x-y" a fancy section name
or is it the expression ".foo = x - y"?  It isn't possible for a
single lookahead parser to decide, so the answer in this case is
that it's a section name.  This is the reason why everyone writes
linker script assignment expressions with lots of white-space.

However, there are many places where the parser knows for sure that an
expression is expected.  Those could be written without whitespace
given the first change to ldlex.l below.  Unfortunately, that runs
into a lookahead problem.  Optional expressions at the end of an
output section statement require the parser to look ahead one token in
expression context.  For this example from standard scripts
  .interp             : { *(.interp) }
  .note.gnu.build-id  : { *(.note.gnu.build-id) }
at the end of the .interp closing brace, the parser is looking for
a possible memspec, phdr, fill or even an optional comma.  The next
token is a NAME, but in expression context that NAME now doesn't
include '-' as a valid char.  So the lookahead NAME is
".note.gnu.build" with an unexpected "-id" syntax error before the
colon.  The rest of the patch involving ldlex_backup arranges to
discard that NAME token so that it will be rescanned in the proper
script context.

	* ldgram.y (section): Call ldlex_backup.  Remove empty action.
	* ldlex.h (ldlex_backup): Declare.
	* ldlex.l (<EXPRESSION>NAME): Don't use NOCFILENAMECHAR set of
	chars, use SYMBOLNAMECHAR.
	(ldlex_backup): New function.


-- 
Alan Modra
Australia Development Lab, IBM

Patch

diff --git a/ld/ldgram.y b/ld/ldgram.y
index b0a4619bb3a..08dc110f3da 100644
--- a/ld/ldgram.y
+++ b/ld/ldgram.y
@@ -1071,11 +1071,15 @@  section:	NAME		{ ldlex_expression(); }
 		'}' { ldlex_popstate (); ldlex_expression (); }
 		memspec_opt memspec_at_opt phdr_opt fill_opt
 		{
+		  if (yychar == NAME)
+		    {
+		      yyclearin;
+		      ldlex_backup ();
+		    }
 		  ldlex_popstate ();
 		  lang_leave_output_section_statement ($18, $15, $17, $16);
 		}
 		opt_comma
-		{}
 	|	OVERLAY
 			{ ldlex_expression (); }
 		opt_exp_without_type opt_nocrossrefs opt_at opt_subalign
@@ -1089,6 +1093,11 @@  section:	NAME		{ ldlex_expression(); }
 			{ ldlex_popstate (); ldlex_expression (); }
 		memspec_opt memspec_at_opt phdr_opt fill_opt
 			{
+			  if (yychar == NAME)
+			    {
+			      yyclearin;
+			      ldlex_backup ();
+			    }
 			  ldlex_popstate ();
 			  lang_leave_overlay ($5, (int) $4,
 					      $16, $13, $15, $14);
diff --git a/ld/ldlex.h b/ld/ldlex.h
index b0101028321..d9b36ea2702 100644
--- a/ld/ldlex.h
+++ b/ld/ldlex.h
@@ -191,6 +191,7 @@  extern void ldlex_defsym (void);
 extern void ldlex_expression (void);
 extern void ldlex_both (void);
 extern void ldlex_popstate (void);
+extern void ldlex_backup (void);
 extern const char* ldlex_filename (void);
 
 /* In lexsup.c.  */
diff --git a/ld/ldlex.l b/ld/ldlex.l
index 237892c0ec3..7652e8d2a29 100644
--- a/ld/ldlex.l
+++ b/ld/ldlex.l
@@ -385,7 +385,7 @@  V_IDENTIFIER [*?.$_a-zA-Z\[\]\-\!\^\\]([*?.$_a-zA-Z0-9\[\]\-\!\^\\]|::)*
 				  yylval.name = xstrdup (yytext + 2);
 				  return LNAME;
 				}
-<EXPRESSION>{SYMBOLNAMECHAR1}{NOCFILENAMECHAR}* {
+<EXPRESSION>{SYMBOLNAMECHAR1}{SYMBOLNAMECHAR}* {
 				  yylval.name = xstrdup (yytext);
 				  return NAME;
 				}
@@ -636,6 +636,16 @@  ldlex_popstate (void)
   yy_start = *(--state_stack_p);
 }
 
+/* In cases where the parser needs to look ahead and the context
+   changes from expression to script or vice-versa, throw away a
+   NAME.  What constitutes a NAME depends on context.  */
+
+void
+ldlex_backup (void)
+{
+  yyless (0);
+}
+
 /* Return the current file name, or the previous file if no file is
    current.  */