On Mon, Feb 27, 2017 at 3:54 PM, Takashi Iwai tiwai@suse.de wrote:
On Mon, 27 Feb 2017 14:49:00 +0100, Linus Walleij wrote:
If CONFIG_SND_VERBOSE_PROCFS is enabled I got a bunch of nasty messages from duplicate entries when registering the DAIs like this:
[ 3.144805] entry->name == "info" [ 3.148132] entry->name == "info" [ 3.151519] entry->name == "hw_params" [ 3.155273] entry->name == "sw_params" [ 3.159027] entry->name == "status" [ 3.164001] entry->name == "prealloc" [ 3.167663] entry->name == "prealloc_max" [ 3.173034] entry->name == "prealloc" [ 3.176696] ------------[ cut here ]------------ [ 3.181335] WARNING: CPU: 0 PID: 95 at ../fs/proc/generic.c:346 proc_register+0xf0/0x110 [ 3.189422] proc_dir_entry 'sub0/prealloc' already registered (...) [ 3.414642] entry->name == "prealloc_max" [ 3.418670] ------------[ cut here ]------------ [ 3.423309] WARNING: CPU: 0 PID: 95 at ../fs/proc/generic.c:346 proc_register+0xf0/0x110 [ 3.431396] proc_dir_entry 'sub0/prealloc_max' already registered (...) [ 3.658020] snd-soc-mop500 soc:sound: ab8500-codec-dai.0 <-> 80124000.msp mapping ok [ 3.665893] entry->name == "info" [ 3.669219] entry->name == "info" [ 3.672546] entry->name == "hw_params" [ 3.676300] entry->name == "sw_params" [ 3.680053] entry->name == "status" [ 3.684906] entry->name == "prealloc" [ 3.688568] entry->name == "prealloc_max" [ 3.693939] entry->name == "prealloc" [ 3.697601] ------------[ cut here ]------------ [ 3.702239] WARNING: CPU: 0 PID: 95 at ../fs/proc/generic.c:346 proc_register+0xf0/0x110 [ 3.710296] proc_dir_entry 'sub0/prealloc' already registered (...) [ 3.936645] entry->name == "prealloc_max" [ 3.940673] ------------[ cut here ]------------ [ 3.945281] WARNING: CPU: 0 PID: 95 at ../fs/proc/generic.c:346 proc_register+0xf0/0x110 [ 3.953369] proc_dir_entry 'sub0/prealloc_max' already registered [ 4.180023] snd-soc-mop500 soc:sound: ab8500-codec-dai.1 <-> 80125000.msp mapping ok
What I can't seem to figure out is whether the DAI in my card (sound/soc/ux500/mop500.c) is misconfigured some way, or if CONFIG_SND_VERBOSE_PROCFS is not handling this the right way?
If I take out the config option everything seems to work smoothly but I guess it should be possible to have this on?
The warnings are about PCM substream procfs, and the conflicting ones are created by snd_pcm_lib_preallocate_pages*() calls.
Is it a deferred probe? I can imagine that some driver forgot to clean up and left the preallocated stuff. For example, soc-generic-dmaengine-pcm.c calls snd_pcm_lib_preallocate_pages*() but has no counterpart for freeing it.
Ah! I bet a million to one you are spot on on this bug, it seems intuitively correct: this is indeed a deferred probe.
I will try to fix it properly by patching the generic dmaengine driver.
Yours, Linus Walleij