Skip to contentSkip to conversation

WARN_ON() is buggy for 32 bit systems

DISCUSSION Dan Carpenter 3 participants 5 messages Jan 26, 2022 – Jan 27, 2022 PowerPC

lore
Share & tools
lore
#0 lore raw

Hi Michael,

Commit e432fe97f3e5 ("powerpc/bug: Cast to unsigned long before passing to inline asm") breaks WARN_ON() for 32 bit systems.

arch/powerpc/include/asm/bug.h

   109  #define WARN_ON(x) ({                                           \
   110          bool __ret_warn_on = false;                             \
   111          do {                                                    \
   112                  if (__builtin_constant_p((x))) {                \
   113                          if (!(x))                               \
   114                                  break;                          \
   115                          __WARN();                               \
   116                          __ret_warn_on = true;                   \
   117                  } else {                                        \
   118                          __label__ __label_warn_on;              \
   119                                                                  \
   120                          WARN_ENTRY(PPC_TLNEI " %4, 0",          \
   121                                     BUGFLAG_WARNING | BUGFLAG_TAINT(TAINT_WARN), \
   122                                     __label_warn_on,             \
   123                                     "r" ((__force long)(x)));    \
                                                         ^^^^

If the code is "if (WARN_ON(some_u64)) {" then the cast to long will truncate away the high bits so it's wrong. (Or at least that's how it works on x86, I'm working on a work around for Smatch to be able to parse this WARN_ON(). I don't know anything about PowerPC.)

   124                          break;                                  \
   125  __label_warn_on:                                                \
   126                          __ret_warn_on = true;                   \
   127                  }                                               \
   128          } while (0);                                            \
   129          unlikely(__ret_warn_on);                                \

130 })

regards,
dan carpenter

#1 lore raw

Hi Dan,

Le 26/01/2022 à 12:56, Dan Carpenter a écrit :

Quoted 5 lines

Hi Michael,

Commit e432fe97f3e5 ("powerpc/bug: Cast to unsigned long before passing to inline asm") breaks WARN_ON() for 32 bit systems.

I think you missed commit db87a7199229 ("powerpc/bug: Remove specific powerpc BUG_ON() and WARN_ON() on PPC32")

Quoted 18 lines

arch/powerpc/include/asm/bug.h

    109  #define WARN_ON(x) ({                                           \
    110          bool __ret_warn_on = false;                             \
    111          do {                                                    \
    112                  if (__builtin_constant_p((x))) {                \
    113                          if (!(x))                               \
    114                                  break;                          \
    115                          __WARN();                               \
    116                          __ret_warn_on = true;                   \
    117                  } else {                                        \
    118                          __label__ __label_warn_on;              \
    119                                                                  \
    120                          WARN_ENTRY(PPC_TLNEI " %4, 0",          \
    121                                     BUGFLAG_WARNING | BUGFLAG_TAINT(TAINT_WARN), \
    122                                     __label_warn_on,             \
    123                                     "r" ((__force long)(x)));    \
                                                          ^^^^
Quoted 4 lines

If the code is "if (WARN_ON(some_u64)) {" then the cast to long will truncate away the high bits so it's wrong. (Or at least that's how it works on x86, I'm working on a work around for Smatch to be able to parse this WARN_ON(). I don't know anything about PowerPC.)

The code is enclosed in a #ifdef CONFIG_PPC64, it is not used for PPC32:

/arch/powerpc/include/asm/bug.h
99 #ifdef CONFIG_PPC64
100 #define BUG_ON(x) do { \
...
109 #define WARN_ON(x) ({ \
...

  132  #define HAVE_ARCH_BUG_ON
  133  #define HAVE_ARCH_WARN_ON
  134  #endif

Christophe

Quoted 12 lines
    124                          break;                                  \
    125  __label_warn_on:                                                \
    126                          __ret_warn_on = true;                   \
    127                  }                                               \
    128          } while (0);                                            \
    129          unlikely(__ret_warn_on);                                \
    130  })

regards,
dan carpenter

#2 lore raw
Quoted 4 lines · Christophe Leroy

The code is enclosed in a #ifdef CONFIG_PPC64, it is not used for PPC32:

/arch/powerpc/include/asm/bug.h
99 #ifdef CONFIG_PPC64

Ah...

You know, life would be a lot easier for me personally if we added an #ifndef __CHECKER__ as well... I can't compile PowerPC code so I can't test a patch like that.

regards,
dan carpenter

#3 lore raw
Quoted 11 lines · Dan Carpenter
Quoted 4 lines · Christophe Leroy

The code is enclosed in a #ifdef CONFIG_PPC64, it is not used for PPC32:

/arch/powerpc/include/asm/bug.h
99 #ifdef CONFIG_PPC64

Ah...

You know, life would be a lot easier for me personally if we added an #ifndef __CHECKER__ as well... I can't compile PowerPC code so I can't test a patch like that.

Ubuntu & Fedora both have cross compilers packaged, or there's cross compilers on kernel.org. But I assume you mean you'd rather not bother compiling for powerpc, which is fair enough.

Do you mean something like below?

I'm not sure about that, as it would prevent sparse from checking the actual BUG_ON code we're using, vs the generic version which we never use on 64-bit. Is there a smatch specific macro we could check?

cheers

arch/powerpc/include/asm/bug.h +1 -1
index 02c08d1492f8..5cbfe9d8232d 100644
@@ -96,7 +96,7 @@ __label_warn_on:						\
 	break;							\
 } while (0)
 
-#ifdef CONFIG_PPC64
+#if defined(CONFIG_PPC64) && !defined(__CHECKER__)
 #define BUG_ON(x) do {						\
 	if (__builtin_constant_p(x)) {				\
 		if (x)						\
#4 lore raw
Quoted 13 lines · Michael Ellerman
Quoted 11 lines · Dan Carpenter
Quoted 4 lines · Christophe Leroy

The code is enclosed in a #ifdef CONFIG_PPC64, it is not used for PPC32:

/arch/powerpc/include/asm/bug.h
99 #ifdef CONFIG_PPC64

Ah...

You know, life would be a lot easier for me personally if we added an #ifndef __CHECKER__ as well... I can't compile PowerPC code so I can't test a patch like that.

Quoted 5 lines

Ubuntu & Fedora both have cross compilers packaged, or there's cross compilers on kernel.org. But I assume you mean you'd rather not bother compiling for powerpc, which is fair enough.

Do you mean something like below?

Yes, please.

Quoted 4 lines

I'm not sure about that, as it would prevent sparse from checking the actual BUG_ON code we're using, vs the generic version which we never use on 64-bit. Is there a smatch specific macro we could check?

There isn't a Smatch define. This shouldn't affect Sparse at all unless there was a bug in the WARN_ON() macro.

regards,
dan carpenter