[FFmpeg-devel] [libav-devel] [PATCH] dv: fix valgrind use of uninitialised value warnings.

Ronald S. Bultje rsbultje at gmail.com
Mon Jul 4 18:46:12 CEST 2011


Hi,

2011/7/4 Clément Bœsch <ubitux at gmail.com>:
> From: Clément Bœsch <clement.boesch at smartjog.com>
>
> ---
> Hi,
>
> As the subject says, the patch fixes valgrind warnings I expect with a DV100
> file. I'm not sure this is a false positive from valgrind thought, and
> unfortunately the sample I use can't be shared.
>
> Here is an extract of expected warnings:
>
>    ==8399== Use of uninitialised value of size 4
>    ==8399==    at 0x82800A4: dv_decode_ac (dv.c:413)
>    ==8399==    by 0x82840EB: dv_decode_video_segment (dv.c:550)
>    ==8399==    by 0x8482A85: avcodec_default_execute (utils.c:440)
>    ==8399==    by 0x8280BB2: dvvideo_decode_frame (dv.c:1098)
>    ==8399==    by 0x8483611: avcodec_decode_video2 (utils.c:747)
>    ==8399==    by 0x804CB59: output_packet (ffmpeg.c:1611)
>    ==8399==    by 0x808D2A1: transcode.constprop.24 (ffmpeg.c:2781)
>    ==8399==    by 0x8087A75: main (ffmpeg.c:4583)
>    ==8399==
>    ==8399== Use of uninitialised value of size 4
>    ==8399==    at 0x828011F: dv_decode_ac (dv.c:410)
>    ==8399==    by 0x82840EB: dv_decode_video_segment (dv.c:550)
>    ==8399==    by 0x8482A85: avcodec_default_execute (utils.c:440)
>    ==8399==    by 0x8280BB2: dvvideo_decode_frame (dv.c:1098)
>    ==8399==    by 0x8483611: avcodec_decode_video2 (utils.c:747)
>    ==8399==    by 0x804CB59: output_packet (ffmpeg.c:1611)
>    ==8399==    by 0x808D2A1: transcode.constprop.24 (ffmpeg.c:2781)
>    ==8399==    by 0x8087A75: main (ffmpeg.c:4583)
>    ==8399==
>    ==8399== Use of uninitialised value of size 4e=00:00:01.06 bitrate=2804.1kbits/s
>    ==8399==    at 0x8280111: dv_decode_ac (dv.c:408)
>    ==8399==    by 0x8284356: dv_decode_video_segment (dv.c:572)
>    ==8399==    by 0x8482A85: avcodec_default_execute (utils.c:440)
>    ==8399==    by 0x8280BB2: dvvideo_decode_frame (dv.c:1098)
>    ==8399==    by 0x8483611: avcodec_decode_video2 (utils.c:747)
>    ==8399==    by 0x804CB59: output_packet (ffmpeg.c:1611)
>    ==8399==    by 0x808D2A1: transcode.constprop.24 (ffmpeg.c:2781)
>    ==8399==    by 0x8087A75: main (ffmpeg.c:4583)
>    ==8399==
>
> Note: it is not a regression (the issue is present since the first DV100
> support was added).
>
> ---
>  libavcodec/dv.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/libavcodec/dv.c b/libavcodec/dv.c
> index d6c49c8..f426d73 100644
> --- a/libavcodec/dv.c
> +++ b/libavcodec/dv.c
> @@ -476,8 +476,8 @@ static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
>     GetBitContext gb;
>     BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
>     LOCAL_ALIGNED_16(DCTELEM, sblock, [5*DV_MAX_BPM], [64]);
> -    LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [80 + 4]); /* allow some slack */
> -    LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5 * 80 + 4]); /* allow some slack */
> +    LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [80 + 4])     = {0}; /* allow some slack */
> +    LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5 * 80 + 4]) = {0}; /* allow some slack */

This is a little redundant, but it's true that each input buffer needs
zero padding and this one doesn't. The easiest solution, I think, is
to add zero padding (plus add the put_flush_bits()?) at the end after
the written bytes (FF_INPUT_BUFFER_PADDING_SIZE or so), s/4/that/ also
above) in bit_copy().

Ronald


More information about the ffmpeg-devel mailing list