Re: IIO channel type for status/error flag indicators

[Jonathan Cameron <jic23@kernel.org>]

On Thu, 28 Mar 2019 18:38:21 +0000
"Popa, Stefan Serban" <StefanSerban.Popa@analog.com> wrote:

> Hi Jonathan,
> 
> Thank you for taking the time to answer.
> 
> These devices are mostly used in automotive/aerospace industries where they
> usually require a continuous stream of data even if an error occurs. This
> is why we cannot just "drop the record" :).
> So, should we maybe think of a way of covering flags channel types?

Propose an interface and we can certainly talk about it :)  Keep in mind
that it needs to be lightweight for anything not making use of it and
that I can see this absorbing a lot of namespace / id space (particularly
in potential event codes) so keep that in mind as well.

I'd argue that those automotive / aerospace devices should certainly
cope with missing data, but perhaps that's wishful thinking.

Anyhow, I'm not against this in principle but think it's a non trivial
exercise so prepare yourself for a big job and quite a few rounds of
interface descriptions!

Jonathan

> 
> Regards,
> -Stefan 
> 
> On Du, 2019-03-24 at 18:00 +0000, Jonathan Cameron wrote:
> > 
> > 
> > On Wed, 20 Mar 2019 14:53:50 +0000
> > "Popa, Stefan Serban" <StefanSerban.Popa@analog.com> wrote:
> >   
> > > 
> > > Hello,
> > > 
> > > We are currently working on a new adis16495 IMU driver which is an
> > > upgrade
> > > from the adis16480 family: https://www.analog.com/media/en/technical-do
> > > cume
> > > ntation/data-sheets/ADIS16495.pdf.
> > > 
> > > This new chip supports a feature called "Burst Read Function" (page 13
> > > in
> > > the datasheet). The burst read function (BRF) provides a method for
> > > reading
> > > a batch of data (status, temperature, gyroscopes, accelerometers, time
> > > stamp/data counter, and CRC code), which does not require a stall
> > > time between each 16-bit segment and only requires one command
> > > on the DIN line to initiate.
> > > 
> > > Most of the data read in this way can be attributed to a type of
> > > channel:
> > > IIO_TEMP, IIO_ANGL_VEL, IIO_ACCEL, etc. However, there is no equivalent
> > > for
> > > the status and CRC. The status register provides various error flags
> > > such
> > > as spi communication error, sensor failure, sync error, etc (Table 18
> > > in
> > > the datasheet). This information together with the CRC error should be
> > > exposed to the user space. What is the best way to do it?  
> > Ok, So this is not exactly unusual.  The issue has always been defining
> > a remotely generic userspace ABI.
> >   
> > > 
> > > 
> > > The most obvious way, but not necessarily the correct one, would be to
> > > add
> > > a new channel type called something like IIO_STATUS or IIO_FLAG. Is
> > > this
> > > acceptable?  
> > No unfortunately. What does generic userspace do with it?
> > 
> > Part of the problem is we don't have a channel type to cover flags in
> > general (if we had digital inputs packed into bytes we would at least
> > be slightly better off).
> >   
> > > 
> > > 
> > > A more or less similar burst read function has been previously
> > > implemented
> > > as part of the adis16400 driver. Although a burst read will also
> > > produce
> > > diagnostic status data, it was ignored in the driver implementation.  
> > Yup :)  Couldn't figure out how to do it at the time.
> > 
> > Normally these flags represent error conditions (if they map to events
> > in IIO then put them out like that).  The problem has always been that
> > Linux doesn't actually have generic simple error event handling.
> > There is RAS handling for servers, but who runs it on embedded boards
> > except possibly for some form of EDAC.
> >   
> > > 
> > > 
> > > Looking forward to any suggestion.  
> > I'm not against us having meta data channels, but they pretty much need
> > to be as tightly defined as any other channel.  The other side issue
> > here is it's a new 'huge' space.  However, I'm not sure you are in
> > that territory here.  Looks to me like status really means error.
> > If you get an error, mostly it's game over.  If you get a CRC error
> > and want to check it, then drop the record and spit out a message.
> > 
> > So I'm a little unconvinced as yet that there is anything we actually
> > can use in this status message.
> > 
> > Jonathan
> > 
> >   
> > > 
> > > 
> > > -Stefa