[alsa-devel] floating point exception triggered by speaker-test on pc-speaker
Hello,
I have hit a strange bug. To reproduce:
1. Build snd-pcsp.ko.
2. alsamixer -c pcsp, set the base frequency to 37286 Hz
3. $ speaker-test -r 22050 -P 4 -D plughw:pcsp
speaker-test 1.0.18
Playback device is plughw:pcsp Stream parameters are 22050Hz, S16_LE, 1 channels Using 16 octaves of pink noise Rate set to 22050Hz (requested 22050Hz) Buffer size range from 37 to 38756 Period size range from 18 to 19379 Using max buffer size 38756 Periods = 4 was set period_size = 0 was set buffer_size = 38756 0 - Front Left Floating point exception
I.e., as you see, period size somehow became 0. While debugging this with printfs, I noticed that the interval for periods in snd_pcm_hw_param_set_near() starts with 4, but doesn't include it (i.e., is open).
Also, I found that snd_interval_refine_first() and snd_interval_refine_last(), when given an open interval, refine it to something like (n n+1). This is obviously wrong, as there are no integer points in this interval. Changing i->max++ and i->min-- there to adjust the value by 2 did eliminate the floating point exception, but replaced it with "Unable to set nperiods 4 for playback: Invalid argument". So there must be something else also wrong, but I can't pinpoint it and give you a ready-made patch :(
At Sun, 23 Nov 2008 14:30:24 +0500, Alexander E. Patrakov wrote:
Hello,
I have hit a strange bug. To reproduce:
Build snd-pcsp.ko.
alsamixer -c pcsp, set the base frequency to 37286 Hz
$ speaker-test -r 22050 -P 4 -D plughw:pcsp
speaker-test 1.0.18
Playback device is plughw:pcsp Stream parameters are 22050Hz, S16_LE, 1 channels Using 16 octaves of pink noise Rate set to 22050Hz (requested 22050Hz) Buffer size range from 37 to 38756 Period size range from 18 to 19379 Using max buffer size 38756 Periods = 4 was set period_size = 0 was set buffer_size = 38756 0 - Front Left Floating point exception
I.e., as you see, period size somehow became 0. While debugging this with printfs, I noticed that the interval for periods in snd_pcm_hw_param_set_near() starts with 4, but doesn't include it (i.e., is open).
Also, I found that snd_interval_refine_first() and snd_interval_refine_last(), when given an open interval, refine it to something like (n n+1). This is obviously wrong, as there are no integer points in this interval. Changing i->max++ and i->min-- there to adjust the value by 2 did eliminate the floating point exception, but replaced it with "Unable to set nperiods 4 for playback: Invalid argument". So there must be something else also wrong, but I can't pinpoint it and give you a ready-made patch :(
Thanks for reporting. The problem is that speaker-test examines the parameters before calling snd_pcm_hw_params(). After calling, these are set properly (even with pcsp driver :)
The fix is committed now.
Takashi
participants (2)
-
Alexander E. Patrakov -
Takashi Iwai