Re: [RFT PATCH v2 2/3] hwmon: (spd5118) Use spd5118 specific read/write operations

[Guenter Roeck <linux@roeck-us.net>]

On 6/18/24 17:50, Thomas Weißschuh wrote:
> On 2024-06-18 17:23:44+0000, Guenter Roeck wrote:
>> On 6/18/24 16:39, Paul Menzel wrote:
>>> [Cc: +Heiner]
>>>
>>>
>>> Dear Armin,
>>>
>>>
>>> Am 19.06.24 um 01:28 schrieb Armin Wolf:
>>>> Am 19.06.24 um 00:28 schrieb Wolfram Sang:
>>>>
>>>>>> to 86 degrees C. If that doesn't work, we'll be really out of luck
>>>>>> with that controller (or at least I don't have an idea what else to try).
>>>>>
>>>>> Try CCing Heiner Kallweit for ideas about the i801 controller.
>>>
>>>> i am not Heiner Kallweit, but i found something interesting in
>>>> commit ba9ad2af7019 ("i2c: i801: Fix I2C Block Read on 8-Series/C220 and later").
>>>>
>>>> Basically, it seems that the i802 i2c controller indeed features a SPD write disable bit which blocks all writes for slave addresses 0x50-0x57.
>>>>
>>>> Does the i801 i2c controller driver print something like "SPD Write Disable is set" during boot?
>>>
>>> Nice find. Yes, it does:
>>>
>>
>> Yes, definitely. I didn't have any recent datasheets, so I missed that flag.
>> Oh well :-(.
>>
>>>       [    5.462605] i801_smbus 0000:00:1f.4: SPD Write Disable is set
>>>       [    5.468399] i801_smbus 0000:00:1f.4: SMBus using PCI interrupt
>>>
>>
>> Bummer. That explains the problem. It means that the BIOS effectively
>> blocks reading the eeprom on your system (because that would require writing
>> the page register), as well as changing temperature limits. That is really
>> annoying, but there is nothing we can do about it. Maybe the BIOS has a
>> configuration flag to enable or disable write protect, but I doubt it.
> 
> What about using 16bit addressing mode?
> 
>      Alternatively, at initial power on, the host can set the Table 112, “MR11” [3] = ‘1’ to address the entire 1024 bytes of
>      non-volatile memory with 2 bytes of address and hence not required to go through page selection to address entire
>      non-volatile memory.
> 
> regmap-i2c allows 16bit addresses when I2C_FUNC_SMBUS_I2C_BLOCK is supported,
> which to me looks like it should be the case on i801 for ICH5.
> 

Good idea, but it doesn't work. I can get write operations with
16-bit register addresses to work even on piix4, but read operations
require writing a 16-bit register address followed by byte reads (see
regmap_i2c_smbus_i2c_read_reg16). Unfortunately, spd5118 devices
don't auto-increment the address on byte read operations, meaning
each byte read returns data from address 0x00 (i.e., it returns
0x51). Try "i2cdump -y -f 0 0x50 c" and you'll see what I mean.
Maybe there is a way around it, but I have not found it.

On top of that, configuring 16-bit mode requires a write operation
into the page register, and that is blocked.

Thanks,
Guenter