[alsa-devel] Duplicate configuration nodes
Hi,
Since I upgraded from alsa-lib 1.0.20 to 1.0.21 my configuration is not parsed correctly anymore:
$ aplay test.wav ALSA lib conf.c:4639:(snd_config_expand) Expand error (walk): File exists Playing WAVE 'test.wav' : Signed 16 bit Little Endian, Rate 44100 Hz, Mono
As a result of this error, ALSA seems to ignore the second half of HDA-Intel.conf, including the iec958 definition.
I traced the "File exists" error back to snd_config_add() returning -EEXIST, while evaluating the "strings" node at line 36 of cards/HDA-Intel.conf. I verified that at that point the src node indeed has four children: two duplicates of each child node of the original strings array.
I'm not exactly sure what's happening here, but I do see that the HDA-Intel.conf file (and all of its included files) are actually read *twice* by aplay. With alsa-lib 1.0.20 aplay only reads these files once.
The problem seems to be related to commit 96da0c842d14b40ce8e37726b259229634b3aa21 ("Fix driver conf parsing in snd_config_hook_load_for_all_cards()"), which subtly changes the logic when the same driver is encountered for the second time. I have two cards using the HDA-Intel config (one analog and one digital). For the first card, the config is loaded without issues. But for the second card, snd_config_search(root, fdriver, &n) succeeds but snd_config_get_string(n, &driver) fails. In 1.0.20 this results in a "goto __err", skipping the actual config load. But in 1.0.21 this results in the else-branch being chosen (driver = fdriver) and next snd_config_hook_load being called.
I know too little of the entire config loading process to be sure what exactly is going wrong here, but a assume that the second call to snd_config_hook_load for the same configaration file somehow eventually results in the duplicate node error in snd_config_add().
I've verified that reverting 96da0c842d14b40ce8e37726b259229634b3aa21 fixes the problem, or at least the symptoms. :-)
Please let me know if you need extra information, of if you'd like me to do some more testing.
Cheers, Jochem Wichers Hoeth
participants (1)
-
Jochem Wichers Hoeth