Hi Simon,

arno at natisbad.org (Arnaud Ebalard) writes:

Regarding your problem, I first started by booting current version of my
driver on the Duo v2 w/o touching chip registers (only clock reference
value). This way, I inherit the values installed by (NETGEAR's) u-boot:

$ for k in fan* pwm* ; do echo -n "$k:" ; echo `cat $k `; done
fan1_alarm:0
fan1_div:2
fan1_fault:0
fan1_input:1807
fan1_target:1807
pwm1:221
pwm1_enable:2        /* closed-loop, i.e. config is done via set_cnt */
pwm1_mode:1          /* PWM mode */ 

If G762 datasheet is correct, NETGEAR's u-boot alters fan divisor ('01'
i.e. 2 instead of '00' i.e. 1). It also changes the default mode from
linear to pwm and sets set_cnt to 34 (255-221). 

Then, I grabbed the GPL sources (5.3.7) of NETGEAR's u-boot and found
the following in a boot function:

 /* Turn G76x(FAN controller, i2c address 0x3e) on.
  * The FAN_SET_CNT register's offset is 0x0.
  * Set [1300(rpm) == 65% == 5a(FAN_SET_CNT)] as default.
  */
 unsigned char byte=0x5a;
 if(i2c_write(0, 0x3e, 0x0, 1, &byte, 1) != 0) 
   puts ("Error writing the i2c chip : G76x(Fan controller).\n");

AFAICT, this is the only location in the whole source tarball where 0x3e
reg is modified in an i2c function. 


Then, I grabbed the u-boot source code from NETGEAR for ReadyNAS 102
(Armada-370 based NAS, also using a G762) and it has the following:

 /* Turn G76x(FAN controller, i2c address 0x3e) on.
  * The FAN_SET_CNT register's offset is 0x0.
  * Set [1300(rpm) == 65% == 5a(FAN_SET_CNT)] as default.
  */
 MV_U8 byte=0x5a;
 if(i2c_write(0x3e, 0x0, 1, &byte, 1) != 0)
 	puts ("Error writing the i2c chip : G76x(Fan controller).\n");
 /* Tune the FAN_STARTV */
 if (i2c_read(0x3e, 0x5, 1, &byte, 1) != 0)
 	puts ("Error reading the i2c chip : G76x(Fan controller).\n");
 byte |= 0x3;
 if(i2c_write(0x3e, 0x5, 1, &byte, 1) != 0)
 	puts ("Error writing the i2c chip : G76x(Fan controller).\n");
 
In the end, it does not seem NETGEAR's u-boot does some specific things
except setting a non-zero value to SET_CNT register for the fan to start
rotating.

BTW, if you wonder why some attributes reported at the beginning have
different values than the ones in the datasheet (0x5a set in u-boot,
reading 0x22) although NETGEAR's u-boot code does not seem to tamper
those, I don't have the answer. 

I was not happy with previous result and did some additional tests.
Between each tests, I did a full power off (halt and removal of power
cable) of the device instead of a simple reboot. With a power off, I
get the following:

fan1_alarm:0
fan1_div:1
fan1_fault:0
fan1_input:1365
fan1_target:1365
pwm1:0
pwm1_enable:2
pwm1_mode:0

Note that fan1_target reflects the fact that set_cnt is 165 (i.e. 255 -
0x5a), which is exactly what NETGEAR's u-boot sets for this reg.

Now, if I do a simple restart, there is some hysteresis which explains
the weird difference in my first test.

Another message follows w/ additional answers to your problem.

Cheers,

a+