On Thu, Apr 30, 2015 at 7:09 AM, Martin Pieuchot <[email protected]> wrote:

> On 24/04/15(Fri) 20:48, David Higgs wrote:
> > This is the final patch in the series.
> >
> > Utilize the pending flags and report callback for their intended purpose
> - to process async behavior.
> >
> > Apply splusb() to ensure report callbacks can't fire before their data
> structures have been properly updated.  This only needs to happen in
> upd_refresh(); all other calls to upd_request_children() are from a report
> callback.
>
> This is some good work.  I don't think your patch #6 makes sense without
> #7, so I merged them.
>

Sounds good, thanks for all the help.

I don't think you need a "pending" flags for the sensors, since what you
> are querying are the reports.  You basically do not want to send to
> requests for the same report at once.
>

In general the code tries to avoid requesting reports more than once, but
this prevents sensor dependencies from causing some nasty problems.  In the
example below, I don't want the update of C to re-update A and trigger an
infinite loop by requesting B.

Sensor A (report 1) -> Sensor B (report 2) -> Sensor C (report 1)

The alternative is to allow inconsistent sensor values.  If the value of B
affects the value of C, I would rather perform a 2nd request for report 1
once I have the value of B, than to put garbage/bogus values into C.

I prefer to keep sensor values consistent, to ensure nobody's sensorsd
starts to panic and because I doubt many real devices have this kind of
annoying HID descriptor layout.


> I'd also take the simpler approach of not deactivating all sensors if
> you fail to *submit* a transfer.  When such thing happens it's either
> because the device/bus is going away or because we cannot allocate
> memory.
>

This sounds fine for the bus/device case or if M_NOWAIT would trigger a
panic.  How about if the usb device itself runs out of memory/buffers; will
the rest of the system stay happy?  Again, I prefer to err on the side of
consistency and invalidate child sensors so sensorsd isn't looking at stale
values.


> I tweaked your diff below, included some comments and your copyright.
> This is untested but I hope you'll correct/improve it.
>

I'll take a look over the next week.


> I'm also curious, which new reports/sensors do you want to add?
>

I haven't looked at this in some months.  My consumer-grade device had
several useful sensors that aren't currently handled.  There are some
"internal" reports (CapacityMode, AtRate) that aren't suitable as sensors,
but might fix other sensor values being mis-reported.  And there are
interactive things that can be done via reports - silencing the audible
alarm, requesting self-tests - that would be good as sysctls or ioctls
someday.

I'll be happy to look at lsusb output if anyone has a device in mind.

Thanks.

--david

Reply via email to