Patchwork audio: fix integer overflow expression

login
register
mail settings
Submitter Peter Maydell
Date May 31, 2011, 5:28 p.m.
Message ID <1306862938-13431-1-git-send-email-peter.maydell@linaro.org>
Download mbox | patch
Permalink /patch/98067/
State New
Headers show

Comments

Peter Maydell - May 31, 2011, 5:28 p.m.
From: Juha Riihimäki <juha.riihimaki@nokia.com>

Fix an integer overflow that can happen for signed 32 bit types
when using FLOAT_MIXENG. (Note that at the moment this is only true
when using the MacOSX coreaudio audio driver.)

Signed-off-by: Juha Riihimäki <juha.riihimaki@nokia.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
---
I'm trying to get random patches out of my patch-stack and upstream.
This one looks "obviously correct" but it only kicks in for MacOSX
and coreaudio, and I don't have access to that platform to test myself,
so treat my reviewed-by accordingly.

This has actually been posted here before, last year:
 http://patchwork.ozlabs.org/patch/48703/

 audio/mixeng_template.h |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
Peter Maydell - May 31, 2011, 6:24 p.m.
This definitely had malc on the cc list when I sent it; as far
as I can tell the mailing list server ate that cc...

-- PMM

On 31 May 2011 18:28, Peter Maydell <peter.maydell@linaro.org> wrote:
> From: Juha Riihimäki <juha.riihimaki@nokia.com>
>
> Fix an integer overflow that can happen for signed 32 bit types
> when using FLOAT_MIXENG. (Note that at the moment this is only true
> when using the MacOSX coreaudio audio driver.)
>
> Signed-off-by: Juha Riihimäki <juha.riihimaki@nokia.com>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> ---
> I'm trying to get random patches out of my patch-stack and upstream.
> This one looks "obviously correct" but it only kicks in for MacOSX
> and coreaudio, and I don't have access to that platform to test myself,
> so treat my reviewed-by accordingly.
>
> This has actually been posted here before, last year:
>  http://patchwork.ozlabs.org/patch/48703/
>
>  audio/mixeng_template.h |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/audio/mixeng_template.h b/audio/mixeng_template.h
> index a2d0ef8..e01da0a 100644
> --- a/audio/mixeng_template.h
> +++ b/audio/mixeng_template.h
> @@ -46,7 +46,7 @@ static mixeng_real inline glue (conv_, ET) (IN_T v)
>  #endif
>  #else  /* !RECIPROCAL */
>  #ifdef SIGNED
> -    return nv / (mixeng_real) (IN_MAX - IN_MIN);
> +    return nv / (mixeng_real) ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN);
>  #else
>     return (nv - HALF) / (mixeng_real) IN_MAX;
>  #endif
> @@ -63,7 +63,7 @@ static IN_T inline glue (clip_, ET) (mixeng_real v)
>     }
>
>  #ifdef SIGNED
> -    return ENDIAN_CONVERT ((IN_T) (v * (IN_MAX - IN_MIN)));
> +    return ENDIAN_CONVERT ((IN_T) (v * ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN)));
>  #else
>     return ENDIAN_CONVERT ((IN_T) ((v * IN_MAX) + HALF));
>  #endif
> --
> 1.7.1
>
>
>
malc - May 31, 2011, 6:30 p.m.
On Tue, 31 May 2011, Peter Maydell wrote:

> From: Juha Riihim?ki <juha.riihimaki@nokia.com>
> 
> Fix an integer overflow that can happen for signed 32 bit types
> when using FLOAT_MIXENG. (Note that at the moment this is only true
> when using the MacOSX coreaudio audio driver.)
> 
> Signed-off-by: Juha Riihim?ki <juha.riihimaki@nokia.com>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> ---
> I'm trying to get random patches out of my patch-stack and upstream.
> This one looks "obviously correct" but it only kicks in for MacOSX
> and coreaudio, and I don't have access to that platform to test myself,
> so treat my reviewed-by accordingly.
> 
> This has actually been posted here before, last year:
>  http://patchwork.ozlabs.org/patch/48703/

One nit, only one of the casts is really needed, pick one and i'll apply
it.

> 
>  audio/mixeng_template.h |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/audio/mixeng_template.h b/audio/mixeng_template.h
> index a2d0ef8..e01da0a 100644
> --- a/audio/mixeng_template.h
> +++ b/audio/mixeng_template.h
> @@ -46,7 +46,7 @@ static mixeng_real inline glue (conv_, ET) (IN_T v)
>  #endif
>  #else  /* !RECIPROCAL */
>  #ifdef SIGNED
> -    return nv / (mixeng_real) (IN_MAX - IN_MIN);
> +    return nv / (mixeng_real) ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN);
>  #else
>      return (nv - HALF) / (mixeng_real) IN_MAX;
>  #endif
> @@ -63,7 +63,7 @@ static IN_T inline glue (clip_, ET) (mixeng_real v)
>      }
>  
>  #ifdef SIGNED
> -    return ENDIAN_CONVERT ((IN_T) (v * (IN_MAX - IN_MIN)));
> +    return ENDIAN_CONVERT ((IN_T) (v * ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN)));
>  #else
>      return ENDIAN_CONVERT ((IN_T) ((v * IN_MAX) + HALF));
>  #endif
>
Andreas Färber - May 31, 2011, 6:52 p.m.
Am 31.05.2011 um 19:28 schrieb Peter Maydell:

> From: Juha Riihimäki <juha.riihimaki@nokia.com>
>
> Fix an integer overflow that can happen for signed 32 bit types
> when using FLOAT_MIXENG. (Note that at the moment this is only true
> when using the MacOSX coreaudio audio driver.)
>

Missing Riku's SoB compared to the link you gave below!

> Signed-off-by: Juha Riihimäki <juha.riihimaki@nokia.com>
> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> ---
> I'm trying to get random patches out of my patch-stack and upstream.
> This one looks "obviously correct" but it only kicks in for MacOSX
> and coreaudio, and I don't have access to that platform to test  
> myself,
> so treat my reviewed-by accordingly.

The warnings I get on v10.5 ppc64 are as follows:

   CC    audio/mixeng.o
In file included from /Users/andreas/QEMU/qemu/audio/mixeng.c:112:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h: In function  
‘conv_natural_int32_t’:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h:49: warning: integer  
overflow in expression
/Users/andreas/QEMU/qemu/audio/mixeng_template.h: In function  
‘clip_natural_int32_t’:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h:66: warning: integer  
overflow in expression
In file included from /Users/andreas/QEMU/qemu/audio/mixeng.c:117:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h: In function  
‘conv_swap_int32_t’:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h:49: warning: integer  
overflow in expression
/Users/andreas/QEMU/qemu/audio/mixeng_template.h: In function  
‘clip_swap_int32_t’:
/Users/andreas/QEMU/qemu/audio/mixeng_template.h:66: warning: integer  
overflow in expression

I confirm that this patch silences these. Will test the new patch  
right away.

> This has actually been posted here before, last year:
> http://patchwork.ozlabs.org/patch/48703/

That patch clearly shows that dumping a large and unrelated patch  
series onto the mailing list without cc'ing people and never following  
them up is not a winning strategy.

Andreas

> audio/mixeng_template.h |    4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/audio/mixeng_template.h b/audio/mixeng_template.h
> index a2d0ef8..e01da0a 100644
> --- a/audio/mixeng_template.h
> +++ b/audio/mixeng_template.h
> @@ -46,7 +46,7 @@ static mixeng_real inline glue (conv_, ET) (IN_T v)
> #endif
> #else  /* !RECIPROCAL */
> #ifdef SIGNED
> -    return nv / (mixeng_real) (IN_MAX - IN_MIN);
> +    return nv / (mixeng_real) ((mixeng_real)IN_MAX -  
> (mixeng_real)IN_MIN);
> #else
>     return (nv - HALF) / (mixeng_real) IN_MAX;
> #endif
> @@ -63,7 +63,7 @@ static IN_T inline glue (clip_, ET) (mixeng_real v)
>     }
>
> #ifdef SIGNED
> -    return ENDIAN_CONVERT ((IN_T) (v * (IN_MAX - IN_MIN)));
> +    return ENDIAN_CONVERT ((IN_T) (v * ((mixeng_real)IN_MAX -  
> (mixeng_real)IN_MIN)));
> #else
>     return ENDIAN_CONVERT ((IN_T) ((v * IN_MAX) + HALF));
> #endif
> -- 
> 1.7.1
>

Patch

diff --git a/audio/mixeng_template.h b/audio/mixeng_template.h
index a2d0ef8..e01da0a 100644
--- a/audio/mixeng_template.h
+++ b/audio/mixeng_template.h
@@ -46,7 +46,7 @@  static mixeng_real inline glue (conv_, ET) (IN_T v)
 #endif
 #else  /* !RECIPROCAL */
 #ifdef SIGNED
-    return nv / (mixeng_real) (IN_MAX - IN_MIN);
+    return nv / (mixeng_real) ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN);
 #else
     return (nv - HALF) / (mixeng_real) IN_MAX;
 #endif
@@ -63,7 +63,7 @@  static IN_T inline glue (clip_, ET) (mixeng_real v)
     }
 
 #ifdef SIGNED
-    return ENDIAN_CONVERT ((IN_T) (v * (IN_MAX - IN_MIN)));
+    return ENDIAN_CONVERT ((IN_T) (v * ((mixeng_real)IN_MAX - (mixeng_real)IN_MIN)));
 #else
     return ENDIAN_CONVERT ((IN_T) ((v * IN_MAX) + HALF));
 #endif