Hi Guenter,
On Wed, 25 Jun 2014 13:42:26 -0700, Guenter Roeck wrote:
> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
> ---
> CHANGES | 1 +
> prog/detect/sensors-detect | 68 ++++++++++++++++++++++++++++++++++++++++++++++
> 2 files changed, 69 insertions(+)
>
> diff --git a/CHANGES b/CHANGES
> index 3390942..a16e7be 100644
> --- a/CHANGES
> +++ b/CHANGES
> @@ -13,6 +13,7 @@ SVN HEAD
> Add detection of ITE IT8620E and IT8623E
> Add detection of TMP441, TMP442, LM95233, LM95234,
> and LM95235
> + Add detection of NCT7802Y
>
> 3.3.5 "Happy Birthday Beddy" (2014-01-22)
> libsensors: Improve documentation of two functions
> diff --git a/prog/detect/sensors-detect b/prog/detect/sensors-detect
> index df9a408..159d2cb 100755
> --- a/prog/detect/sensors-detect
> +++ b/prog/detect/sensors-detect
> @@ -733,6 +733,11 @@ use vars qw(@i2c_adapter_names);
> i2c_addrs => [0x2c..0x2f],
> i2c_detect => sub { w83795_detect(@_); },
> }, {
> + name => "Nuvoton NCT7802Y",
> + driver => "to-be-written",
> + i2c_addrs => [0x28..0x2f],
> + i2c_detect => sub { nct7802_detect(@_); },
> + }, {
> name => "Winbond W83627HF",
> driver => "use-isa-instead",
> i2c_addrs => [0x28..0x2f],
> @@ -5483,6 +5488,69 @@ sub w83795_detect
> }
>
> # Registers used:
> +# 0x00: bank selection (Bank 0, 1)
> +# 0x40, 0x80, 0xc0: bank selection (Bank 1)
> +# 0x01, 0x41, 0x81, 0xc1: PECI control register 1 (Bank 1)
> +# 0x02, 0x42, 0x82, 0xc2: PECI control register 2 (Bank 1)
> +# 0x03, 0x43, 0x83, 0xc3: PECI control register 3 (Bank 1)
> +# 0xfd: Vendor ID (Bank 0)
> +# 0xfe: Device ID (Bank 0)
> +# 0xff: Device Revision (Bank 0)
> +sub nct7802_detect
> +{
> + my ($bank, $reg, $pc1, $pc2, $pc3);
> + my ($file, $addr) = @_;
> +
> + $bank = i2c_smbus_read_byte_data($file, 0x00);
> + return unless ($bank == 0x00 or $bank == 0x01);
> +
> + # we can not do much if bank 1 is selected
> + if ($bank == 0x01) {
> + # We can not use bit masks since all bits
> + # (including reserved bits) can be set.
> + $pc1 = i2c_smbus_read_byte_data($file, 0x01);
> + $pc2 = i2c_smbus_read_byte_data($file, 0x02);
> + $pc3 = i2c_smbus_read_byte_data($file, 0x03);
> +
> + # Register values repeat every 64 registers on bank 1.
> + # Check the first four registers of each group.
> + for (my $i = 0x40; $i <= 0xc0; $i += 0x40) {
> + $reg = i2c_smbus_read_byte_data($file, $i);
> + return if $reg != 0x01;
> + $reg = i2c_smbus_read_byte_data($file, $i + 1);
> + return if $reg != $pc1;
> + $reg = i2c_smbus_read_byte_data($file, $i + 2);
> + return if $reg != $pc2;
> + $reg = i2c_smbus_read_byte_data($file, $i + 3);
> + return if $reg != $pc3;
> + }
> + return 3;
> + }
Bah. I discussed the need to have the identification registers in
non-banked registers with Winbond/Nuvoton engineers years ago, and it
looked like they did listen to me with the W83795G/ADG, but now they
are doing it all wrong again :-(
The above is simply too weak, I'm afraid. Chips cycling over 8, 16, 32
or 64 register addresses are legions. Just from my register dump
collection, the LM80 and the VIA 686A chips would both be detected by
the loop above. There may be more, in particular if Nuvoton keeps using
the same strategy for future chips.
So I would say, just skip the detection if the chip is in bank 1. Let's
hope the BIOS will leave the chip in bank 0. When we have a driver for
that chip, we'll make sure to leave the chip in bank 0 too.
> +
> + # bank 0, can support real chip detection
> +
> + $reg = i2c_smbus_read_byte_data($file, 0xfd);
> + return unless ($reg == 0x50);
Unneeded parentheses.
> +
> + $reg = i2c_smbus_read_byte_data($file, 0xfe);
> + return unless $reg == 0xc3;
> +
> + $reg = i2c_smbus_read_byte_data($file, 0xff);
> + return unless ($reg & 0xf0) == 0x20;
> +
> + $reg = i2c_smbus_read_byte_data($file, 0x05);
> + return unless ($reg & 0x1f) == 0x00;
> +
> + $reg = i2c_smbus_read_byte_data($file, 0x08);
> + return unless ($reg & 0x3f) == 0x00;
> +
> + $reg = i2c_smbus_read_byte_data($file, 0x0f);
> + return unless ($reg & 0x3f) == 0x00;
You did not say above, what these last 3 registers were. Also I am
surprised because checking these is in contradiction with your statement
above that "we can not use bit masks since all bits (including reserved
bits) can be set."
> +
> + return 8;
> +}
> +
> +# Registers used:
> # 0x48: Full I2C Address
> # 0x4e: Vendor ID byte selection
> # 0x4f: Vendor ID
--
Jean Delvare
SUSE L3 Support
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
