Hi All,
The cc list is based on who turns up a lot in my in box so please
forward to any other interested parties!
It's time we updated our todo list and started moves towards
leaving staging. Looking at it the todo is way out of date
so I'll just start from scratch.
First purely administrative change is to push elements of this
TODO down close to the drivers. There are now way to many to
document sensibly in one place. I propose the following structure
iio/TODO with stuff that applies to all drivers.
iio/accel/TODO for general accel stuff.
iio/accel/TODO-lis3l02dq etc for individual drivers.
Anyhow we'll leave drivers for now. Lets consider what
we might have in top level TODO
Obvious and still required.
1) Review.
2) Testing.
Big sections:
* Core
( ida cleanups, though that's not sensible to put in the TODO file.
Basically it's a case of rolling all 'device id' uses of ida's into
a central library with a single lock so as to avoid the huge amount of
boiler plate code.)
* Buffering
1) Replace or get significant review of sw-ring. As I've mentioned many
times I really don't like my code.
2) More options.
3) Documentation. Arnd suggested a man page given unusual nature of the chrdev.
Our main docs probably need a thorough read over and possibly updates as well.
* Triggering
Post the locking / registration fix set, not sure we want to do more in here?
* Events
Post making the fixes Arnd suggested and chasing down allocation / freeing issues
as a result of refactoring in my recent RFC.
1) Event codes need rethink. Michael already has a device where he ran out of space.
Perhaps we just make the code 64 bit right now. It actually costs us almost nothing
given padding of the event structure.
2) Replace the event handling code. It's simplistic and single user. Is this an issue?
3) Consider matching event structure format with input's evdev?
* sysfs attribute names.
1) Consider dropping 'compatibility' with hwmon to allow more consistency. in -> voltage
for example.
2) Have a prefix to specify direction. So in_voltageX_raw out_voltageX_raw. Michael
brought this up. If we are going to do it, now is the time. We may need a
'compatibility mode' to smooth the transition. Note however that we don't want to
carry that mode out of staging. (do we also need an inout option?)
The next big discussion is how to propose a patch set moving out of staging. This is
hopefully where we'll start to get more feedback.
I propose the following:
1) Move the namespace of key exported bits and bobs to iio_st* (for staging)
2) Add a prefix to drivers in kconfig (as we move them);
IIO_ seems the obvious choice.
That leaves us clear to lift code across. How about a series that looks like:
1) core sysfs only infrastructure + docs.
2) A few example drivers
---review period---
3) Event support + docs
4) A few example drivers
---review period---
5) Core buffer support + docs
6) Hardware buffer example (sca3000 given I have one).
7) Software buffer example (kfifo first as it's easier to review).
8) A few example drivers.
---review period---
9) Sets of clean drivers
continue till we run out.
Note some drivers will probably be in staging for quite
a large period - ultimately some need a complete rewrite.
As we go along I suggest we try and keep the staging tree in sync with
any changes. At the end, I'd like to drop the core from the staging
tree if possible and just have all remaining drivers use the new core
implementation whether they are in staging or not.
The only other option that really makes sense is to do the events
and buffering in the opposite order.
So we need to figure out:
a) are we happy with this order.
b) which drivers are we taking at each stage.
c) how to post the changes (linux-iio first then lkml after a week?)
For the which drivers, the set I will personally carry are (listed by where
they are updated in the above tree).
2) max1363, adis16400, tsl2563
4) max1363, tsl2563 (adis16400 doesn't actually have support for events
in tree)
6) sca3000
8) max1363, sysfs trigger.
I'd certainly like to take a few of our other nice drivers along for the
ride but I can't test them. I'm not taking the lis3l02dq for now despite
it being one of my core test drivers, because it will cause issues given
the misc/lis3 driver.
So what do people want to carry through the process?
Mostly it will involve stripping elements out and then slowly building things
back up again as the support becomes available. One important
point I'd like to do is remove the IIO_CHAN macro and do things
explicitly. Partly that macro is becoming restrictive and confusing
+ it will make for a much cleaner build up of support.
All comments welcome!
Once a consensus has formed *cross fingers*, I'll do a clean
version to post more widely so others are kept in the loop.
Jonathan
--
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
