Re: [PATCH] Detect and warn on atomic_inc/atomic_dec wrappingaround

From: Ingo Molnar
Date: Thu Apr 30 2009 - 06:47:53 EST



* Nikanth Karthikesan <knikanth@xxxxxxxxxx> wrote:

> > As a generic debug helper this is not appropriate i think -
> > counts can easily have a meaning when going negative as well.
> > (we have no signed-atomic primitives)
>
> This doesn't warn when it becomes negative/positive from zero, but
> only when it wraps around^W^Woverflows, trying to add past INT_MAX
> or subtract from INT_MIN.

ah! The small difference between INT_MAX and UINT_MAX.

Sure, this debug check makes a lot of sense. -mm even had such
stuff, many years ago? Never went upstream.

> > > static inline void atomic_inc(atomic_t *v)
> > > {
> > > +#if defined(CONFIG_ENABLE_WARN_ATOMIC_INC_WRAP)
> > > + WARN_ON(atomic_add_unless(v, 1, INT_MAX) == 0);
> > > +#else
> > > asm volatile(LOCK_PREFIX "incl %0"
> > >
> > > : "+m" (v->counter));
> > >
> > > +#endif
> > > }
> >
> > also looks a bit ugly - this ugly #ifdef would spread into every
> > architecture.
> >
> > If we want to restrict atomic_t value ranges like that then the
> > clean solution would be to add generic wrappers doing the debug
> > (once, in generic code), and renaming the arch primitives to
> > raw_atomic_inc() (etc), doing the lowlevel bits cleanly.
> >
>
> Here is a patch which does it this way.
>
> Thanks
> Nikanth
>
> Detect and warn on atomic_inc/atomic_dec overflow.
>
> Add a debug option to detect and warn when the 32-bit atomic_t overflows
> during atomic_inc and atomic_dec.
>
> diff --git a/arch/x86/include/asm/atomic_32.h b/arch/x86/include/asm/atomic_32.h
> index 85b46fb..c6a17bb 100644
> --- a/arch/x86/include/asm/atomic_32.h
> +++ b/arch/x86/include/asm/atomic_32.h
> @@ -78,24 +78,24 @@ static inline int atomic_sub_and_test(int i, atomic_t *v)
> }
>
> /**
> - * atomic_inc - increment atomic variable
> + * raw_atomic_inc - increment atomic variable
> * @v: pointer of type atomic_t
> *
> * Atomically increments @v by 1.
> */
> -static inline void atomic_inc(atomic_t *v)
> +static inline void raw_atomic_inc(atomic_t *v)
> {
> asm volatile(LOCK_PREFIX "incl %0"
> : "+m" (v->counter));
> }
>
> /**
> - * atomic_dec - decrement atomic variable
> + * raw_atomic_dec - decrement atomic variable
> * @v: pointer of type atomic_t
> *
> * Atomically decrements @v by 1.
> */
> -static inline void atomic_dec(atomic_t *v)
> +static inline void raw_atomic_dec(atomic_t *v)
> {
> asm volatile(LOCK_PREFIX "decl %0"
> : "+m" (v->counter));
> diff --git a/arch/x86/include/asm/atomic_64.h b/arch/x86/include/asm/atomic_64.h
> index 8c21731..1183b85 100644
> --- a/arch/x86/include/asm/atomic_64.h
> +++ b/arch/x86/include/asm/atomic_64.h
> @@ -77,12 +77,12 @@ static inline int atomic_sub_and_test(int i, atomic_t *v)
> }
>
> /**
> - * atomic_inc - increment atomic variable
> + * raw_atomic_inc - increment atomic variable
> * @v: pointer of type atomic_t
> *
> * Atomically increments @v by 1.
> */
> -static inline void atomic_inc(atomic_t *v)
> +static inline void raw_atomic_inc(atomic_t *v)
> {
> asm volatile(LOCK_PREFIX "incl %0"
> : "=m" (v->counter)
> @@ -90,12 +90,12 @@ static inline void atomic_inc(atomic_t *v)
> }
>
> /**
> - * atomic_dec - decrement atomic variable
> + * raw_atomic_dec - decrement atomic variable
> * @v: pointer of type atomic_t
> *
> * Atomically decrements @v by 1.
> */
> -static inline void atomic_dec(atomic_t *v)
> +static inline void raw_atomic_dec(atomic_t *v)
> {
> asm volatile(LOCK_PREFIX "decl %0"
> : "=m" (v->counter)
> diff --git a/include/asm-generic/atomic.h b/include/asm-generic/atomic.h
> index 7abdaa9..6eda22b 100644
> --- a/include/asm-generic/atomic.h
> +++ b/include/asm-generic/atomic.h
> @@ -4,15 +4,52 @@
> * Copyright (C) 2005 Silicon Graphics, Inc.
> * Christoph Lameter
> *
> - * Allows to provide arch independent atomic definitions without the need to
> - * edit all arch specific atomic.h files.
> */
>
> +#include <linux/kernel.h>
> #include <asm/types.h>
> +#include <asm/bug.h>
> +
> +#if defined(CONFIG_ENABLE_WARN_ATOMIC_INC_WRAP)

#ifdef CONFIG_ENABLE_WARN_ATOMIC_INC_WRAP

> +
> +/**
> + * atomic_inc - increment atomic variable
> + * @v: pointer of type atomic_t
> + *
> + * Atomically increments @v by 1.
> + * Prints a warning if it wraps around.
> + */
> +static inline void atomic_inc(atomic_t *v)
> +{
> + WARN_ON(atomic_add_unless(v, 1, INT_MAX) == 0);
> +}
> +
> +/**
> + * atomic_dec - decrement atomic variable
> + * @v: pointer of type atomic_t
> + *
> + * Atomically decrements @v by 1.
> + * Prints a warning if it wraps around.
> + */
> +static inline void atomic_dec(atomic_t *v)
> +{
> + WARN_ON(atomic_add_unless(v, -1, INT_MIN) == 0);
> +}
> +
> +#else
> +
> +#define atomic_inc(v) raw_atomic_inc(v)
> +#define atomic_dec(v) raw_atomic_dec(v)

Please turn these into proper, type aware inline functions.

> +config ENABLE_WARN_ATOMIC_INC_WRAP
> + bool "Enable warning on atomic_inc()/atomic_dec() wrap"
> + default y
> + help
> + Enable printing a warning when atomic_inc() or atomic_dec()
> + operation wraps around the 32-bit value.
> +

this needs HAVE_ARCH_DEBUG_ATOMIC added to arch/x86/Kconfig, and a
depends on HAVE_ARCH_DEBUG_ATOMIC. Otherwise this wont build very
well on non-x86 when enabled, right?

With those small details fixed:

Acked-by: Ingo Molnar <mingo@xxxxxxx>

Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/