The old code was based on a DHT11 datasheet which specifies a measurement
range of 20%-90% RH. Turns out the sensor actually reports values outside
this range, so we should support it as far as possible.
Reported-by: Edward Attfield <edward@attfield.ca>
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
The equivalent of both of these are now done via macro magic when
the relevant register calls are made. The actual structure
elements will shortly go away.
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: Lars-Peter Clausen <lars@metafoo.de>
The DHT22 (AM2302) datasheet specifies that the LOW start pulse should not
exceed 20ms. However, observations with an oscilloscope of an RPi Model 2B
(rev 1.1) communicating with a DHT22 sensor showed that the driver was
consistently sending start pulses longer than 20ms:
Kernel 4.7.10-v7+ (n=132):
Minimum pulse length: 20.20ms
Maximum: 29.84ms
Mean: 24.96ms
StDev: 2.82ms
Sensor response rate: 100%
Read success rate: 76%
On kernel 4.8, the start pulse was so long that the sensor would not even
respond 97% of the time:
Kernel 4.8.16-v7+ (n=100):
Minimum pulse length: 30.4ms
Maximum: 74.4ms
Mean: 39.3ms
StDev: 10.2ms
Sensor response rate: 3%
Read success rate: 3%
The driver would return ETIMEDOUT and write log messages like this:
[ 51.430987] dht11 dht11@0: Only 1 signal edges detected
[ 66.311019] dht11 dht11@0: Only 0 signal edges detected
Replacing msleep(18) with usleep_range(18000, 20000) made the pulse length
sane again and restored responsiveness:
Kernel 4.8.16-v7+ with usleep_range (n=123):
Minimum pulse length: 18.16ms
Maximum: 20.20ms
Mean: 19.85ms
StDev: 0.51ms
Sensor response rate: 100%
Read success rate: 84%
Cc: stable@vger.kernel.org
Signed-off-by: John Brooks <john@fastquake.com>
Reviewed-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
* Unify log messages
* Add more DEBUG messages
Apparently this driver is working unreliably on some platforms that I can't
test. Therefore I want an easy way for bug reporters to provide useful
information without making the driver too chatty by default.
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
Wall time obtained from ktime_get_real_ns is susceptible to sudden jumps due to
user setting the time or due to NTP. Boot time is constantly increasing time
better suited for comparing two timestamps.
Signed-off-by: Abhilash Jindal <klock.android@gmail.com>
Reviewed-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
The new algorithm uses a 'one size fits em all' threshold, which should
be easier to understand and debug. I believe there are no regressions
compared to the old adaptive threshold algorithm. I don't remember why
I chose the old algorithm when I initially wrote the driver.
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
Instead of guessing where the data starts, we now just try to decode from
every possible start position. This causes no additional overhead if we
properly received the full preamble and only costs a few extra CPU cycles
in the case where the preamble is corrupted. This is much more efficient
than to return an error to userspace and start over again.
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
This cleans up the most ugly workaround in this driver. There are no
functional changes yet in the decoding algorithm, but we improve the
following things:
* Get rid of spurious warning messages on systems with fast HRTIMER.
* If the clock is not fast enough for decoding to work, we give
up immediately.
* In that case we return EAGAIN instead of EIO, so it's easier to
discriminate causes of failure.
Returning EAGAIN is somewhat controversial: It's technically correct
as a faster clock might become available. OTOH once all clocks are
enabled this is a permanent error. There is no ECLOCKTOOSLOW error
code.
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
* add spaces around binary operators in cases where it reduces readability
* align multiline statements around opening parenthesis
Reported-by: Hartmut Knaack <knaack.h@gmx.de>
in Message-ID: <55919E72.3010807@gmx.de>
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
Since setting irq-enabled GPIOs into output state is not supported
by all GPIO controllers, we need to disable the irq while requesting
sensor data. As side effect we lose a tiny bit of functionality:
Some wiring problems can't be concluded from log messages anymore.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Harald Geyer <harald@ccbib.org>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
Make sure that the read function is not interrupted...
Signed-off-by: Richard Weinberger <richard@nod.at>
Acked-by: Harald Geyer <harald@ccbib.org>
Reviewed-by: Sanjeev Sharma <sanjeev_sharma@mentor.com>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
As we access i-1 we must not start with i=0.
Signed-off-by: Richard Weinberger <richard@nod.at>
Acked-by: Hartmut Knaack <knaack.h@gmx.de>
Acked-by: Harald Geyer <harald@ccbib.org>
Reviewed-by: Sanjeev Sharma <sanjeev_sharma@mentor.com>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
This patch removes the .owner field for drivers which use the
platform_driver_register api because this is overriden in
_platform_driver_register.
Signed-off-by: Sanjeev Sharma <Sanjeev_Sharma@mentor.com>
Signed-off-by: Jonathan Cameron <jic23@kernel.org>
Harald Geyer