Re: [RFC 4/4] iio: ina2xx: add SOFTWARE buffer mode using an iio kfifo.

From: Marc Titinger
Date: Mon Nov 16 2015 - 04:37:47 EST


On 14/11/2015 19:44, Jonathan Cameron wrote:
On 12/11/15 10:18, Marc Titinger wrote:
On 10/11/2015 19:23, Lars-Peter Clausen wrote:
On 11/10/2015 05:07 PM, Marc Titinger wrote:
Capture the active scan_elements into a kfifo.
The capture thread will compute the remaining time until the next capture
tick, and do an active wait (udelay).

This will produce a stream of up to fours channels plus a 64bits
timestamps (ns).

# iio_readdev ina226 | od -x
WARNING: High-speed mode not enabled
0000000 042f 0d5a 002e 010c 4be8 4eb4 0013 0000
0000020 0430 0d5a 002e 010c a704 4f3e 0013 0000
0000040 0430 0d5a 002e 010c b477 4fc7 0013 0000
0000060 042f 0d5b 002e 010c 8052 5050 0013 0000
0000100 042f 0d5b 002e 010c 5d92 50d8 0013 0000
0000120 0430 0d5a 002e 010c fa59 515e 0013 0000
0000140 0430 0d5b 002e 010c 95d2 51e5 0013 0000

Signed-off-by: Marc Titinger <mtitinger@xxxxxxxxxxxx>

Hi Lars,


Interesting approach. I think if we are going to due this we want to make
this kind of emulation generic. Have you seen the software trigger and
configfs support patches[1] from Daniel? It kind of achieves the same as you
do, but using hrtimers.


I totally agree, let me have a look on those patches maybe I could
add an active waiting scheme for platforms w/o hrtimers ?

I've no idea if this is a remotely common case any more but in theory
I'd have no objection to such a patch though I would like it to be
a stand alone trigger similar to that used in the patch Daniel has
submitted.


Yes, I've started looking into something like a 'polling' trigger class. I understand that I'd need to support triggered buffer mode in the driver in this case. The current patch is rather simple, but having a generic and documented solution seems a good idea.



---
Ina2xx does not support auto-increment, hence the capture threads sticks
with single register reads instead of regmap_bulk_read.

The proper scales must be applied to those raw register
values, I'm in favor of doing the conversion in userland in a client plugin

Yes, conversion should not be done in kernel space, we don't want to impose
the performance penalty on users which don't need it and you can typically
do it faster in userspace anyway where you have floats and SSE and what not.

for instance a sigrok

Slightly OT, but do you already have some Sigrok IIO support? I have this
scheduled for end of the month, maybe we can align our strategies here and
avoid duplicated work.


How fortunate! I've started some preliminary work like cloning the
demo driver into a skeletton for 'hardware/generic-iio/api.c', adding
the build/ac plumbing, and linking to libiio with the idea of using
iio_info to create a generic enumeration of the iio-context into
sigrok channels.

Now, I'm not familiar with Glib and it might not be my prio until a
couple of weeks, so I'd be super happy to wait for you if you are
keen to do that part :)

What would be the best spot to chat about this ?


Marc.



- Lars

[1] https://lkml.org/lkml/2015/8/10/877


--
To unsubscribe from this list: send the line "unsubscribe linux-iio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/