Marvell Ethernet PHYs can contain a temperature sensor, which the kernel
exports as an HWMON device. The PHY is on an MDIO bus. Add support for
this bus types to libsensors. The sensor is then displayed by sensors(1):
socaipsbus02100000ethernet02188000mdioswitch0mdio01-mdio-1
Adapter: MDIO adapter
temp1: +54.0°C (crit = +100.0°C)
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Add support for sysfs attributes temp[1-*]_min_hyst (already
implemented by drivers adt7x10, lm77 and lm92) and
temp[1-*]_lcrit_hyst (no known users yet.)
While there is no longer a hard limit to the number of sensor of a
given type per chip, I feel a little uncomfortable having no limit at
all on the amount of memory we may try to allocate. Add an arbitrary
safety limit so that a design error or a bug in a hwmon driver can't
result into an insane memory allocation.
Make memory pre-allocation steps depend on the sensor type. Things like
voltages, temperatures or fans are typically plenty, however vid and
chassis intrusion are typically only a few, and there can only be one
beep_enable by design. This saves a small amount of temporary memory
for cheap.
max_subfeatures is computed the first time it is needed, we can do
exactly the same with FEATURE_SIZE. Introduce feature_size as a static
variable so that we don't have to recompute it again and again.
When gathering the attributes of each hwmon chip, libsensors uses a
temporary structure in memory to order and group all the attributes
into features. This temporary structure used to be a single array with
room for every possible attribute/subfeature. While simple, this
approach required to predefine a maximum number of per-type sensor
that could be handled.
In order to get rid of this arbitrary limit, which we hit and had to
raise three times already, I changed the temporary structure to an
array of dynamically allocated per-type subattribute arrays. This lets
us not allocate any memory for types which aren't implemented by a
given chip, and more importantly, this lets us reallocate room for
more attributes of a given type as needed.
I decided to allocate chunks of 8 attributes at a time, as this seemed
a good compromise between two frequent reallocations and
over-provisioning. It could be tweaked if needed.
Icing on the cake, I benchmarked this change on two different systems
and it results in performance gains. The total heap usage as reported
by valgrind is down by 50% on average, with peak memory consumption
(as reported by valgrind's massif) also down by 43% on average. The
total instructions count (as reported by valgrind's callgrind) is down
by 11% on average, with measured execution time also down by a few
percents. Valgrind rocks, BTW.
I have some ideas to optimize the memory allocations further, but I do
not expect such a huge gain from them. They may not even improve peak
memory consumption as massif shows the peak is somewhere else now at
least in some cases.
attributes. These are defined in the standard sysfs interface for quite
some time, and at least three drivers (max6650, lm63 and applesmc)
implement them so we should support them.
git-svn-id: http://lm-sensors.org/svn/lm-sensors/trunk@6029 7894878c-1315-0410-8ee3-d5d059ff63e0
now, but this make it possible to add more feature types without
increasing the code complexity beyond reason. Define constants and use
them to make the code even easier to understand.
If anyone is really worried about the relatively large memory
allocation, one way to improve the situation would be a two-pass
approach. But remember the memory is only needed for a short time...
git-svn-id: http://lm-sensors.org/svn/lm-sensors/trunk@5874 7894878c-1315-0410-8ee3-d5d059ff63e0
class device directory rather than directly in the device directory.
The latter is what all drivers do at the moment, but in the long run
the former is preferred as it prevents attribute name collisions.
git-svn-id: http://lm-sensors.org/svn/lm-sensors/branches/lm-sensors-3.0.0@5093 7894878c-1315-0410-8ee3-d5d059ff63e0
guessing it from how the device identifier looks like. For kernels
<= 2.6.17, we use the bus symlink instead. For kernels <= 2.6.11,
neither symlink exist so we fallback to the old type guessing
approach. It doesn't really matter as all hwmon devices were
i2c devices back then.
git-svn-id: http://lm-sensors.org/svn/lm-sensors/branches/lm-sensors-3.0.0@5092 7894878c-1315-0410-8ee3-d5d059ff63e0
Instead, access sysfs directly, using 3 embedded helper
functions. My motivations for doing this are:
* As far as I know, libsysfs is no longer maintained.
* libsysfs does much more than we need. For example, when asking for a
device attribute list, libsysfs will read the contents and permissions
of all attributes. Not only does this waste CPU cycles per se, but in
the case of hwmon driver it also triggers register reads, which can be
slow for SMBus chips.
* libsysfs enforces the difference between devices and class devices,
while future changes will be easier if we can handle both types alike.
git-svn-id: http://lm-sensors.org/svn/lm-sensors/branches/lm-sensors-3.0.0@5067 7894878c-1315-0410-8ee3-d5d059ff63e0
simply ignore the errors. In debug mode, call sensors_fatal_error().
As a side note, there is certainly room for improvement in the way
errors are reported by libsensors. sensors_fatal_error() is fatal,
and sensors_parse_error() is too specific, so we lack a more general
error reporting function. printf-like formatting for error messages
would also be a good idea.
git-svn-id: http://lm-sensors.org/svn/lm-sensors/branches/lm-sensors-3.0.0@4855 7894878c-1315-0410-8ee3-d5d059ff63e0
Andrew Lunn