after stepping back from the 23 patches, it seems theres too many little
pieces.
So heres a reroll, treating the 5 sensor-types together, since thats
pretty much
how the code treats them now. (ie repeated patterns).
Hopefully it will all look a bit more rational & discussable now (ie 3rd
time's the charm)
This time, patches are against 13-rc4-mm1
1. Yani form change.
a. declare SENSOR_DEVICE_ATTR, not DEVICE_ATTR
b. in device_create_file() calls, pass &dev_attr member of the
sensor-dev-attr type.
c. in sensor-attr-[gs]etter callback-functions,
use to_sensor_dev-attr to get sensor_dev_attr *attr
use attr->index instead of offset
d. include hwmon-sysfs.h, needed for a.
Note that this isnt a global replace. Only those attrs that would
benefit from
having an index have been converted; ie those which are macro-repeated.
2. function renames (preparation for de-macro-ization of getter-setter
callbacks)
avoid name-clashes when we de-macro-ize later.
Currently we have:
423 #define show_and_set_therm(offset) \
424 static ssize_t show_temp##offset##_input(struct device *dev,
struct device_attribute *attr, char *buf 424 ) \
425 { \
537 #define show_and_set_temp(offset) \
538 static ssize_t show_temp##offset##_input(struct device *dev,
struct device_attribute *attr, char *buf 538 ) \
539 { \
these are distinguished purely by the ##offset## in the callback name,
which will go away in next patch, so we get ready by changing temp to therm
in the show_and_set_therm macro.
2nd situation is similar, this time with one of the fan callbacks and a
fan-helper function.
248
249 static ssize_t set_fan_min(struct device *dev, const char *buf,
250 size_t count, int nr)
251 {
302 } \
303 static ssize_t set_fan##offset##_min(struct device *dev, struct
device_attribute *attr, const char *b 303 uf, \
304 size_t count) \
3. convert macro-repeated callback-fns to single declaration
theyre already using attr->index (patch 1)
so this is as close as possible to a white-space patch.
4. callback-fn offset de-skew
temp, therm, fan, pwm callbacks all have an offset skew in the code
which accommodates attribute numbering conventions under
/sys/bus/i2c/devices/9191-6620/ (ie they start at 1)
here, we move that skew into the declaration, and out of the functions
(except for therm, where we simpify from 2 skews to 1)
The declarative skew is clearer, less error-prone, and more efficient.
And for further discussion, as needed...
array initializations
5. refactor SENSOR_DEVICE_ATTR into declaration = initialization-expression
__SENSOR_DEVICE_ATTR() expands to init-expr
this imitates __ATTR in include/linux/device.h,
but is needed cuz of new .index member ( otherwize __ATTR would suffice)
352 #define DEVICE_ATTR(_name,_mode,_show,_store) \
353 struct device_attribute dev_attr_##_name =
__ATTR(_name,_mode,_show,_store)
6. explicit array initializations
a. declare arrays, initialize them with repeated __SENSOR_DEVICE_ATTR's
b. loop over calls to device_create_file()
b could be done separately, but it seems silly -
all the calls need to be touched for a anyway, might as well do so
just once.
and its not like youd declare an array wo planning to use it.
This patch is motivated/justified by comments from GregKH, in msg 11 of:
http://groups-beta.google.com/group/linux.kernel/browse_frm/thread/fa10a47a3c843d08/2a1465afa2f83afe?q=dynamic+sysfs+callbacks&rnum=10&hl=en#2a1465afa2f83afe
> > No, I hate HEAD and TAIL macros. This really isn't buying you much
> > code savings, you could do it yourself with the __ATTR() macro
> > yourself with the same ammount of code I bet...
Hm, which makes me want to go look at trying to convert those attributes
to an array right now...
SO. all 6 patches compile clean, and test ok, modulo the untestable
temp stuff on my soekris.
tfyc,
Jim Cromie
