On 01/27/2015 02:34 PM, Jean Delvare wrote:
On Tue, 27 Jan 2015 12:05:53 -0800, Guenter Roeck wrote:
On Tue, Jan 27, 2015 at 08:54:34PM +0100, Robert Rosengren wrote:
Guenter and Jean,
To sum up, my problems was related my kernel and hardware configuration, and it now works. Many thanks for your input!
However, the values retrieved from hwmon sysfs is not the same as before the regmap patch. Guenter, the byte swap for the regval retrieved by regmap_read. In what order is the bits returned from that function, because it seems as if I disabled that code I get values as I expect (i.e. before the regmap patch).
Trying to understand. Are you saying everything works as expected
if you keep byte_swap set to false ?
That might well be, though it might mean that regmap has a bug
in how it treats i2c word read operations. I'll have to look into it
some more.
Remember that SMBus specifies that the LSB comes first (and i2c-core
implement things that way) while real I2C devices typically send the MSB
first. This has always caused confusion. This is why a lot of drivers
need byte-swapping.
regmap specifies "normal" i2c accesses as MSB first. When SMBus accesses
are used, it specifies LSB first. I guess that means that regmap assumes
the more common case of MSB first, meaning the driver would not need a byte
swap since the chip reports data with MSB first.
I just hope I interpret it correctly this time.
There is also a configuration flag "val_format_endian" which can be set to
REGMAP_ENDIAN_BIG or REGMAP_ENDIAN_LITTLE. Maybe that can be used as well
if needed. Either case I'll wait for the samples before I send an updated
version of the patch.
Guenter
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
