This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] Fix two undefined behaviors in gcc sources


Hi!

--with-build-config=bootstrap-ubsan on i686-linux showed
most often these two issues.

The first one is that 32-bit signed time_t is multiplied by 1000, which
overflows the int type.  The result is then cast to unsigned, so even on
64-bit hosts we don't care about the upper bits.

The other change fixes a badly written portable rotate, which was shitfing
up (correctly) by 0 for i == 0, but for all other i values was shifting up
64-bit value by 64.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2014-03-21  Jakub Jelinek  <jakub@redhat.com>

	* toplev.c (init_local_tick): Avoid signed integer multiplication
	overflow.
	* genautomata.c (reserv_sets_hash_value): Fix rotate idiom, avoid
	shift by first operand's bitsize.

--- gcc/toplev.c.jj	2014-01-08 10:23:22.000000000 +0100
+++ gcc/toplev.c	2014-03-21 11:32:25.929893013 +0100
@@ -261,7 +261,7 @@ init_local_tick (void)
 	struct timeval tv;
 
 	gettimeofday (&tv, NULL);
-	local_tick = tv.tv_sec * 1000 + tv.tv_usec / 1000;
+	local_tick = (unsigned) tv.tv_sec * 1000 + tv.tv_usec / 1000;
       }
 #else
       {
--- gcc/genautomata.c.jj	2014-02-12 08:33:34.000000000 +0100
+++ gcc/genautomata.c	2014-03-21 12:01:14.245727924 +0100
@@ -3494,7 +3494,7 @@ reserv_sets_hash_value (reserv_sets_t re
     {
       reservs_num--;
       hash_value += ((*reserv_ptr >> i)
-		     | (*reserv_ptr << ((sizeof (set_el_t) * CHAR_BIT) & -i)));
+		     | (*reserv_ptr << (((sizeof (set_el_t) * CHAR_BIT) - 1) & -i)));
       i++;
       if (i == sizeof (set_el_t) * CHAR_BIT)
 	i = 0;

	Jakub


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]