Skip to contentSkip to conversation

Is there any word about this bug in gcc ?

DISCUSSION zhengyi 11 participants 17 messages Nov 20, 2007 – Nov 21, 2007

lore
Share & tools
lore
#1 lore raw
Quoted 14 lines · zhengyi

Is there any relevance to the kernel ?

I found the folowing code here:
http://linux.solidot.org/article.pl?sid=07/11/19/0512218&from=rss

-------------------------------------------------------------------
int main( void )
{
int i=2;
if( -10*abs (i-1) == 10*abs(i-1) )
printf ("OMG,-10==10 in linux!\n");
else
printf ("nothing special here\n") ;

Quoted 2 lines

return 0 ;
}

I think no. It is considered a bug in abs(), kernel, of course, doesn't use glibc's abs().

Regards.

#2 lore raw
Quoted 15 lines · WANG Cong
Quoted 14 lines · zhengyi

Is there any relevance to the kernel ?

I found the folowing code here:
http://linux.solidot.org/article.pl?sid=07/11/19/0512218&from=rss

-------------------------------------------------------------------
int main( void )
{
int i=2;
if( -10*abs (i-1) == 10*abs(i-1) )
printf ("OMG,-10==10 in linux!\n");
else
printf ("nothing special here\n") ;

Quoted 6 lines
Quoted 2 lines

return 0 ;
}

I think no. It is considered a bug in abs(), kernel, of course, doesn't use glibc's abs().

Wrong.

abs() is internal to gcc, and the above is optimized out at compile time, so any user of abs() as a function at all is vulnerable.

However, the Linux kernel defines abs() as a macro:

#define abs(x) ({                               \
                int __x = (x);                  \
                (__x < 0) ? -__x : __x;         \
        })

... which means gcc never sees it. So the kernel isn't affected, because it doesn't use *gcc's* abs().

-hpa

#3 lore raw
Quoted 19 lines · H. Peter Anvin
Quoted 18 lines · WANG Cong
Quoted 16 lines · zhengyi

Is there any relevance to the kernel ?

I found the folowing code here:
http://linux.solidot.org/article.pl?sid=07/11/19/0512218&from=rss

-------------------------------------------------------------------
int main( void )
{
int i=2;
if( -10*abs (i-1) == 10*abs(i-1) )
printf ("OMG,-10==10 in linux!\n");
else
printf ("nothing special here\n") ;

return 0 ;
}

Quoted 8 lines
Quoted 3 lines

I think no. It is considered a bug in abs(), kernel, of course, doesn't use glibc's abs().

Wrong.

abs() is internal to gcc, and the above is optimized out at compile time, so any user of abs() as a function at all is vulnerable.

This is an urgent bug, I think.

And you mean abs() is not in glibc, then where is it? Built in gcc? And what's more, why not put it in glibc?

Thanks.

Quoted 10 lines

However, the Linux kernel defines abs() as a macro:

#define abs(x) ({                               \
               int __x = (x);                  \
               (__x < 0) ? -__x : __x;         \
       })

... which means gcc never sees it. So the kernel isn't affected, because it doesn't use *gcc's* abs().

Thanks for clarifying this!

Regards.

#4 lore raw
Quoted 6 lines · WANG Cong

This is an urgent bug, I think.

And you mean abs() is not in glibc, then where is it? Built in gcc? And what's more, why not put it in glibc?

If you need answers to this type of questions, this is not the place for it.

-hpa

#5 lore raw
In reply to WANG Cong
Quoted 24 lines · WANG Cong
Quoted 23 lines · H. Peter Anvin
Quoted 15 lines · WANG Cong
Quoted 14 lines · zhengyi

Is there any relevance to the kernel ?

I found the folowing code here:
http://linux.solidot.org/article.pl?sid=07/11/19/0512218&from=rss

-------------------------------------------------------------------
int main( void )
{
int i=2;
if( -10*abs (i-1) == 10*abs(i-1) )
printf ("OMG,-10==10 in linux!\n");
else
printf ("nothing special here\n") ;

Quoted 5 lines
Quoted 2 lines

return 0 ;
}

I think no. It is considered a bug in abs(), kernel, of course, doesn't use glibc's abs().

Wrong.

Quoted 8 lines
Quoted 2 lines

abs() is internal to gcc, and the above is optimized out at compile time, so any user of abs() as a function at all is vulnerable.

This is an urgent bug, I think.

And you mean abs() is not in glibc, then where is it? Built in gcc? And what's more, why not put it in glibc?

Gcc optimises abs() to use gcc builtin-in abs(). So if we use -fno-builin, we'll get the correct result. That is to say the bug has nothing to do with glibc.

And this bug has been fixed just several days ago.

http://www.nabble.com/-PATCH--Fix-PR34130,-extract_muldiv-broken-t4826688.html

Quoted 15 lines
Quoted 10 lines

However, the Linux kernel defines abs() as a macro:

#define abs(x) ({                               \
               int __x = (x);                  \
               (__x < 0) ? -__x : __x;         \
       })

... which means gcc never sees it. So the kernel isn't affected, because it doesn't use *gcc's* abs().

Thanks for clarifying this!

Regards.

#6 lore raw
Quoted 34 lines · Li Zefan
Quoted 24 lines · WANG Cong
Quoted 23 lines · H. Peter Anvin
Quoted 15 lines · WANG Cong
Quoted 14 lines · zhengyi

Is there any relevance to the kernel ?

I found the folowing code here:
http://linux.solidot.org/article.pl?sid=07/11/19/0512218&from=rss

-------------------------------------------------------------------
int main( void )
{
int i=2;
if( -10*abs (i-1) == 10*abs(i-1) )
printf ("OMG,-10==10 in linux!\n");
else
printf ("nothing special here\n") ;

Quoted 5 lines
Quoted 2 lines

return 0 ;
}

I think no. It is considered a bug in abs(), kernel, of course, doesn't use glibc's abs().

Wrong.

Quoted 8 lines
Quoted 2 lines

abs() is internal to gcc, and the above is optimized out at compile time, so any user of abs() as a function at all is vulnerable.

This is an urgent bug, I think.

And you mean abs() is not in glibc, then where is it? Built in gcc? And what's more, why not put it in glibc?

Quoted 8 lines

Gcc optimises abs() to use gcc builtin-in abs(). So if we use -fno-builin, we'll get the correct result. That is to say the bug has nothing to do with glibc.

And this bug has been fixed just several days ago.

http://www.nabble.com/-PATCH--Fix-PR34130,-extract_muldiv-broken-t4826688.html

Good explanation! Thank you!

#7 lore raw
In reply to WANG Cong

From: WANG Cong <[email protected]>
Date: Tue, 20 Nov 2007 13:39:05 +0800

Quoted 2 lines

And you mean abs() is not in glibc, then where is it? Built in gcc? And what's more, why not put it in glibc?

Because the compiler knows things about the inputs and can thus apply optimizations that a static implementation in glibc that has to handle all forms of inputs cannot.

#8 lore raw
Quoted 4 lines · David Miller

Because the compiler knows things about the inputs and can thus apply optimizations that a static implementation in glibc that has to handle all forms of inputs cannot.

On an unrelated note, I wonder if distros will be treating this with the same level of urgency as security vulnerabilities, especially in light of Shamir's recent note on maths errors.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#9 lore raw
Quoted 9 lines · Herbert Xu
Quoted 3 lines · David Miller

Because the compiler knows things about the inputs and can thus apply optimizations that a static implementation in glibc that has to handle all forms of inputs cannot.

On an unrelated note, I wonder if distros will be treating this with the same level of urgency as security vulnerabilities, especially in light of Shamir's recent note on maths errors.

This one is definitely messy. There is absolutely no way to know what gcc has miscompiled. It looks to me that both gcc 4.2 and 4.3 are affected, any others?

-hpa

#10 lore raw
Quoted 4 lines · H. Peter Anvin

This one is definitely messy. There is absolutely no way to know what gcc has miscompiled. It looks to me that both gcc 4.2 and 4.3 are affected, any others?

I just tested it here and gcc 3.3 is also affected so presumably everything in between is too. Gcc 2.95 is not affected. I don't have the intervening versions to test.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#11 lore raw
Quoted 9 lines · Herbert Xu
Quoted 4 lines · H. Peter Anvin

This one is definitely messy. There is absolutely no way to know what gcc has miscompiled. It looks to me that both gcc 4.2 and 4.3 are affected, any others?

I just tested it here and gcc 3.3 is also affected so presumably everything in between is too. Gcc 2.95 is not affected. I don't have the intervening versions to test.

Fedora 7's 4.1.2-27 is also affected.

--alessandro

"you feel the sweet breath of time
it's whispering, its truth not mine"

(Interpol, 'No I In Threesome')

#12 lore raw
Quoted 7 lines · Alessandro Suardi
Quoted 6 lines · Herbert Xu
Quoted 4 lines · H. Peter Anvin

This one is definitely messy. There is absolutely no way to know what gcc has miscompiled. It looks to me that both gcc 4.2 and 4.3 are affected, any others?

Quoted 5 lines
Quoted 3 lines

I just tested it here and gcc 3.3 is also affected so presumably everything in between is too. Gcc 2.95 is not affected. I don't have the intervening versions to test.

Fedora 7's 4.1.2-27 is also affected.

-m32:
2.7.2.3: /tmp/cc1EO0wg.s:14: Error: suffix or operands invalid for `push' 2.95.3: OK
3.0.4: broken
3.1.1: broken
3.2.3: broken
3.3.5: broken
3.4.6: broken
4.0.3: broken

-m32 and -m64:
gcc Red Hat 4.1.2-33: broken
4.2.2 20070909 (prerelease): broken

Signature · 1 line
Do what you love because life is too short for anything else.
#13 lore raw
In reply to H. Peter Anvin

On 20 Nov 2007, H. Peter Anvin outgrape:

Quoted 2 lines

This one is definitely messy. There is absolutely no way to know what gcc has miscompiled.

Actually, since this only affects abs() calls containing multiplications or divisions by negative constants, you can at least make a pretty good guess as to its prevalence with nothing more than grep (first grep -w for abs, then grepping for "- *[0-9]" and filtering the rest by eye). (This won't catch cases of macros using abs() which are then called with negative constants, but it *will* catch those macros, and one can check the calls to such things by hand. I've done that as well.)

I've grepped all the source on my system (1148 expanded upstream source tarballs or git/cvs/svn trees including the Linux kernel, most of GNOME, and all of KDE and X.org) and found that hits are extremely rare: not as rare as calls to seekdir() and telldir() :) but rare. (Quite a lot of things multiply by negative constants *inside* a call to abs(), but this should be unaffected.)

Certain hits:

./nethack/3.4.3/src/cmd.c: else if(x < -2*abs(y))
./nethack/3.4.3/src/cmd.c: else if(y < -2*abs(x))

Possible hits (I'm not sure what the folder would do with this: the extra level of brackets in the way might affect things but I don't think so):

./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+Pitch]-Src[j]));
./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+1]-Src[j]));

./xmms/modules/projectM-0.94.20/main.c: wave_x_temp=-2*0.4142*(abs(abs(wave_mystery)-.5)-.5);
./xmms/modules/projectM-0.94.20/main.c: wave_x_temp=-2*0.4142*(abs(abs(wave_mystery)-.5)-.5);

None of these affected programs strike me as being exactly system- critical. I think the impact of this bug is probably survivable. :)

I'd build a GCC with the patch and verify that these programs are compiled differently with it, but they look unimportant enough that I'm not really sure I care enough to do it...

Signature · 2 lines
`Some people don't think performance issues are "real bugs", and I think 
such people shouldn't be allowed to program.' --- Linus Torvalds
#14 lore raw
Quoted 6 lines · Nix

I've grepped all the source on my system (1148 expanded upstream source tarballs or git/cvs/svn trees including the Linux kernel, most of GNOME, and all of KDE and X.org) and found that hits are extremely rare: not as rare as calls to seekdir() and telldir() :) but rare. (Quite a lot of things multiply by negative constants *inside* a call to abs(), but this should be unaffected.)

I implemented a different approach: patched gcc with the official fix plus a call to emit a warning (see below), and recompiled the whole LFS LiveCD (see the list of packages at http://wiki.linuxfromscratch.org/livecd/browser/trunk/packages). Only libtheora emitted a warning.

Quoted 4 lines

Certain hits:

./nethack/3.4.3/src/cmd.c: else if(x < -2*abs(y))
./nethack/3.4.3/src/cmd.c: else if(y < -2*abs(x))

Sure, this is a hit, but nethack is not on my LiveCD.

Quoted 6 lines

Possible hits (I'm not sure what the folder would do with this: the extra level of brackets in the way might affect things but I don't think so):

./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+Pitch]-Src[j]));
./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+1]-Src[j]));

This did emit a warning, I have already reported it: https://trac.xiph.org/ticket/1260

Quoted 2 lines

./xmms/modules/projectM-0.94.20/main.c: wave_x_temp=-2*0.4142*(abs(abs(wave_mystery)-.5)-.5);
./xmms/modules/projectM-0.94.20/main.c: wave_x_temp=-2*0.4142*(abs(abs(wave_mystery)-.5)-.5);

Not a hit, probably due to conversions between int and double.

gcc/fold-const.c +3 -0
@@ -6095,6 +6095,9 @@
             }
           break;
         }
+      /* If the constant is negative, we cannot simplify this.  */
+      if (tree_int_cst_sgn (c) == -1)
+        { warning(0, "Unpatched gcc miscompiles this"); break; }
       /* FALLTHROUGH */
     case NEGATE_EXPR:
       if ((t1 = extract_muldiv (op0, c, code, wide_type, strict_overflow_p))
Signature · 1 line
Alexander E. Patrakov
#15 lore raw
Quoted 9 lines · I wrote:
Quoted 6 lines · Nix

Possible hits (I'm not sure what the folder would do with this: the extra level of brackets in the way might affect things but I don't think so):

./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+Pitch]-Src[j]));
./libtheora/libtheora/lib/enc/pp.c: TmpMod = 32 + QValue - 2*(abs(Src[j+1]-Src[j]));

This did emit a warning, I have already reported it: https://trac.xiph.org/ticket/1260

And on IRC, they explained that it is a piece of code that never gets called. So not a hit.

Signature · 1 line
Alexander E. Patrakov
#16 lore raw
Quoted 2 lines · Alexander E. Patrakov

And on IRC, they explained that it is a piece of code that never gets called. So not a hit.

It's still a hit. If the code is never called it should not be there in the first place.

--
Len Sorensen