My previous mail got blocked by mailing lists for some reasons, so
trying once more. I apologize if spamming.
On 01/28/2015 07:18 AM, Robert Rosengren wrote:
On 01/28/2015 05:06 AM, Guenter Roeck wrote:
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 ?
Yes it works as expected, i.e. I can get values from sysfs. The
problems I experienced was actually due to moving from an old
ads7828-implementation where specific i2c addresses where scanned from
the driver, and I first missed to set the correct i2c address for the
new driver in the board info.
>>
>>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
I tested the driver on 3.15 version of the kernel. Having looked in
the git log of regmap.c and regmap-i2c.c, there has been some
endianess-related patches going into the kernel in newer versions.
Maybe that is the reason for me seeing this problem? I will try to
look into that further.
I my test I only disabled following two lines of code
+ if (data->byte_swap)
+ regval = swab16(regval);
BR, Robert
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
