sd_espeak segfaulting with alsa

[Klaus Knopper]

Hi Luke,

On Sat, Jan 10, 2009 at 09:20:00AM +1100, Luke Yelavich wrote:
> On Fri, Jan 09, 2009 at 11:48:00PM EST, Klaus Knopper wrote:
> > Dear list,
> > 
> > For almost a year, I'm experiencing random segfaults in
> > speech-dispatcher like this (dmesg):
> 
> So have I, and thanks to Ubuntu's apport, was able to get this threaded stack 
> trace, see attachment.
> 
> Hope this helps in identifying the problem.

Indeed, it seems to me that in

> #5  0x00007fc262096e1a in snd_pcm_mmap_writei (pcm=0x7fc258017ee4, 
> buffer=0x7fc258024fe8, size=936) at pcm_mmap.c:183
> No locals.
> #6  0x00007fc26386a545 in alsa_play (id=0x23e2900, track=
>       {bits = 16, num_channels = 1, sample_rate = 22050, num_samples = 23751, 
> samples = 0x23d15b0}) at alsa.c:607
>       format = SND_PCM_FORMAT_S16_LE
>       bytes_per_sample = 2
>       num_bytes = <value optimized out>
>       output_samples = (short int *) 0x7fc258024e24
>       real_volume = <value optimized out>
>       err = <value optimized out>
>       ret = <value optimized out>
>       framecount = 470
>       period_size = 470
>       samples_per_period = <value optimized out>
>       silent_samples = <value optimized out>
>       volume_size = 18446744073709551605
>       sr = 22050
>       state = <value optimized out>
>       __PRETTY_FUNCTION__ = "alsa_play"

the volume_size has a much too high number. In fact, it looks like
MAXLONG minus 10, so I guess there is an initialization or overflow
happening somewhere else.

The result is that either malloc fails (if overcommit is set to "off")
or copy functions write over data area limits.

Now I try to find out why volume_size is being calcuated wrongly.

Thanks for sending me this. Can you tell me how the trace as produced?

Regards
-Klaus