Skip to contentSkip to conversation

ALSA drivers for USB audio devices

DISCUSSION Jeffrey W. Baker 3 participants 4 messages Mar 19, 2002

lore
Share & tools
lore
#0 lore raw

Recently on this very list, Paul Barton-Davis proposed that one could build a driver for a USB audio device entirely in user space. I think this is an interesting idea, as I own an Edirol UA-5 USB audio device. Naturally I would like to use it with ALSA to acheive the best performance and lowest latency.

I hope someone on this list can describe how such a driver would work. I am not aware of any user-space drivers for ALSA. If there is one, that would be a nice example. Would the driver take the form of a userspace daemon? How would the ALSA library communicate with the daemon? Or, would the driver instead take the form of a shared library which ALSA libraries would load?

In the daemon case, I have trouble imagining that a userspace implementation would acheive low latency. The client application, say a mixing board program, would have to write some data to the ALSA PCM handle, then wait for the daemon to get scheduled. The latency would be all over the place.

In the case of a library that ALSA loads, that scheduling problem would not exist.

In any case, it seems almost MORE straightforward to write an kernel driver like the other supported output devices use. A 24-bit/96KHz stereo audio stream requires driving the USB bus at nearly full throttle. I doubt a userspace implementation could deal with it.

Thoughts?

-jwb

PS: even at 16-bit/48KHz the USB audio device driver in Linux 2.4.18 taxes the CPU significantly.

_______________________________________________
Alsa-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/alsa-devel

#1 lore raw
Quoted 1 line

Recently on this very list, Paul Barton-Davis proposed that one could

^^^^^^^^^^^^^^^^^
Paul Davis (op.net/~pbd/name.html)

Quoted 2 lines

build a driver for a USB audio device entirely in user space. I think this is an interesting idea, as I own an Edirol UA-5 USB audio device.

Pray that it follows the "tandards" The existing OSS-based driver is full of special case code for just about every device so far supported.

Quoted 9 lines

Naturally I would like to use it with ALSA to acheive the best performance and lowest latency.

I hope someone on this list can describe how such a driver would work. I am not aware of any user-space drivers for ALSA. If there is one, that would be a nice example. Would the driver take the form of a userspace daemon? How would the ALSA library communicate with the daemon? Or, would the driver instead take the form of a shared library which ALSA libraries would load?

there are two approaches:

1) a new low-level ALSA driver that handles interactions with a USB audio device, and presents the ALSA "midlevel" code with the normal *internal* ALSA API

2) a shared library that would be loaded by alsa-lib, based on a description in an asoundrc file. this library would use an existing USB driver that allowed raw "USB packet" I/O - it would format the USB packets in user space based on requests from alsa-lib, and then deliver them to the kernel USB driver.

Quoted 4 lines

In any case, it seems almost MORE straightforward to write an kernel driver like the other supported output devices use. A 24-bit/96KHz stereo audio stream requires driving the USB bus at nearly full throttle. I doubt a userspace implementation could deal with it.

the userspace implementation is only a translation layer between the ALSA PCM API and the USB audio packet standard. it would be an existing USB kernel driver that would take care of streaming the data to and from the h/w. i don't think the user space code would have any problems with this. the kernel driver might, but if so, its a crock anyway, and needs reimplementing. i don't know either way.

i suspect that it would be faster to write the low-level module, but more interesting and potentially more flexible to write the user space library.

--p

_______________________________________________
Alsa-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/alsa-devel

#2 lore raw
Quoted 3 lines · Paul Davis
Quoted 1 line

Recently on this very list, Paul Barton-Davis proposed that one could

^^^^^^^^^^^^^^^^^
Paul Davis (op.net/~pbd/name.html)

                                                ^^^^^^
$ telnet op.net 80
Trying 207.29.195.4...
telnet: connect to address 207.29.195.4: Connection refused
$ telnet www.op.net 80
Trying 207.29.195.25...
Connected to www.op.net.
Escape character is '^]'.

-Dan

Signature · 7 lines
[-] Omae no subete no kichi wa ore no mono da. [-]


_______________________________________________
Alsa-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/alsa-devel
#3 lore raw
In reply to Paul Davis
Quoted 3 lines · Paul Davis
Quoted 1 line

Recently on this very list, Paul Barton-Davis proposed that one could

^^^^^^^^^^^^^^^^^
Paul Davis (op.net/~pbd/name.html)

My mistake. Sorry.

Quoted 6 lines
Quoted 2 lines

build a driver for a USB audio device entirely in user space. I think this is an interesting idea, as I own an Edirol UA-5 USB audio device.

Pray that it follows the "tandards" The existing OSS-based driver is full of special case code for just about every device so far supported.

It supposedly uses some Steinberg driver that supports a number of USB devices, not all from Roland. That gives me hope. However, the ASIO driver doesn't support 24/96 operation, so this may be a device-by-device extension. Duplex operation is limited to 24/48 anyway because of USB bandwidth. I won't care as long as it works on my hardware :)

Quoted 6 lines

there are two approaches:

1) a new low-level ALSA driver that handles interactions with a USB audio device, and presents the ALSA "midlevel" code with the normal *internal* ALSA API

Quoted 6 lines

2) a shared library that would be loaded by alsa-lib, based on a description in an asoundrc file. this library would use an existing USB driver that allowed raw "USB packet" I/O - it would format the USB packets in user space based on requests from alsa-lib, and then deliver them to the kernel USB driver.

The latter sounds more workable -- building packets is userspace is not hard at all, and profiling for performance is more practical.

Quoted 10 lines

the userspace implementation is only a translation layer between the ALSA PCM API and the USB audio packet standard. it would be an existing USB kernel driver that would take care of streaming the data to and from the h/w. i don't think the user space code would have any problems with this. the kernel driver might, but if so, its a crock anyway, and needs reimplementing. i don't know either way.

i suspect that it would be faster to write the low-level module, but more interesting and potentially more flexible to write the user space library.

Probably less duplication of code as well.

-jwb

_______________________________________________
Alsa-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/alsa-devel