Takashi Iwai wrote:

At Sun, 09 Mar 2008 23:56:01 +0100, Pavel Hofman wrote:

...

Hi,

I have most of the ESI Juli driver working, but I have hit a problem with the capture stream. I had to modify some ice1724.c code to allow redefining clock-related functions in the actual card file (Juli has a different clocking scheme), but all the DMA code is left intact, as well as the methods defining the capture stream.

The ice1724 vu-meters detect input ADC data, I can hear correct sound when setting the internal mixer to H/W-In. Yet, when trying to record, I get

arecord -v -D hw:0,0,0 -r 44100 -f S32_LE -c2

.... RIFF$WAVEfmt ... data arecord: pcm_read:1346: read error: File descriptor in bad state

It's -EBADFD and indicates that the PCM state isn't the expected one, in this case, SNDRV_PCM_STATE_PREPARED or RUNNING. The error comes from snd_pcm_capture_ioctl1() in core/pcm_native.c. Not sure what is broken, but this should be the starting point.

Takashi

Takashi,

Thanks for the hint. I have found the following:

For the working capture (spdif input in my case), the function snd_pcm_lib_read1 gets called first with SNDRV_PCM_STATE_PREPARED, and consequently repeatedly with SNDRV_PCM_STATE_RUNNING.

strace arecord : ... write(1, "data\0\0\0\200", 8data) = 8 ioctl(4, 0x800c4151, 0xbfbf5310) = 0 write(1, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 16384) = 16384 ioctl(4, 0x800c4151, 0xbfbf5310) = 0 write(1, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 16384) = 16384 ...

Whereas for the broken analog input, the sequence is

1. snd_pcm_lib_read1 with SNDRV_PCM_STATE_PREPARED - returns 0 2. snd_pcm_lib_read1 with SNDRV_PCM_STATE_SETUP - returns -EBADFD

strace arecord:

write(1, "data\0\0\0\200", 8data) = 8 ioctl(4, 0x800c4151, 0xbfa6cff0) = 0 poll([{fd=4, events=POLLIN|POLLERR|POLLNVAL, revents=POLLIN|POLLERR}], 1, 1000) = 1 ioctl(4, 0x800c4151, 0xbfa6cff0) = -1 EBADFD (File descriptor in bad state) write(2, "arecord: pcm_read:1346: ", 24arecord: pcm_read:1346: ) = 24

The broken input falls to polling - is it somehow blocked?

Thanks a lot for further hints.

Pavel.

Clemens Ladisch wrote:

Pavel Hofman wrote:

...

Whereas for the broken analog input, the sequence is

1. snd_pcm_lib_read1 with SNDRV_PCM_STATE_PREPARED - returns 0
2. snd_pcm_lib_read1 with SNDRV_PCM_STATE_SETUP - returns -EBADFD

I'd guess that your trigger callback fails somehow, bit this is hard to determine without the source code.

Thanks, I will do more testing in the evening. Just in case, I am including the relevant files. I moved all rate-relevant stuff to card-specific routines, redefined in juli.c. The trigger code was not changed at all.

Perhaps those new snd_pcm_hardware structs in juli.c (I do not know the exact meaning of the _bytes_ ones)?

Thanks a lot.

Pavel.