Re: [lm-sensors] [RFT][PATCH 1/2] hwmon: (adm9240) Avoid forward declaration

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

Hi Jean,

On 07/05/2014 11:20 AM, Jean Delvare wrote:
> Hi Guenter,
>
> On Sat, 05 Jul 2014 07:43:45 -0700, Guenter Roeck wrote:
>> On 07/05/2014 06:39 AM, Guenter Roeck wrote:
>>> Something else though that would help: SMBus block commands.
>>> Any idea why this is not currently supported ? I see that
>>> I2C_SMBUS_I2C_BLOCK_DATA is supported, so I2C_SMBUS_BLOCK_DATA
>>> should not be hard. Am I missing something ?
>>
>> I took a stab at that. Guess the main difference to
>> I2C_SMBUS_I2C_BLOCK_DATA is that it needs a separate buffer
>> for the SMBus data block; the 'word' buffer can not be used.
>> Turns out that was quite straightforward to implement.
>
> The main problem I see is that for I2C_SMBUS_BLOCK_DATA reads, the chip
> first returns the number of data bytes (as opposed to
> I2C_SMBUS_I2C_BLOCK_DATA where the controller decides how many bytes it
> wants to read.) There is no way the i2c-stub driver can guess that byte
> count, as it depends on the chip it is supposed to emulate (and might
> even change dynamically, at least in theory.)
>
> We could have limited support for that, but that would require extra
> module parameters to specify the block size for every register offset
> on which SMBus block reads can be attempted. This also assumes that these
> block sizes are static. And as you found out, that may also require
> allocating extra memory for every such register offset.
>
I 'solved' the problem so far by only returning smbus block data after
it was written into a specific command. This way it is all dynamic.

> But another difficulty is also that when SMBus block reads enter the
> game, the usual read/write symmetry tends to disappear. Often the
> registers you read with SMBus block read commands are also readable and
> writable at individual register addresses. i2c-stub has no way to know
> that. Drivers would typically use SMBus block reads for performance
> reason, but byte writes for convenience. So drivers operating on top of
> i2c-stub would get confused in no time.
>
> All these issues have so far convinced me that adding support for SMBus
> block read/writes to i2c-stub wasn't worth it. That being said, if you
> have a specific chip in mind that could be supported easily, I have no
> objection.
>
The one I needed it for right now was lineage-pem; it was useful for me
since I don't have easy access to the HW anymore. Of course, problem with
that is what you pointed out ... with my change in the i2c-stub driver,
the lm93 driver now assumes that it can execute block commands. The only
way I see around that would be a module parameter. That would have to be
one which selects if the smbus block command should be treated similar
to the i2c block command or as individual command blocks. I'll play around
with it some more. I'll send a patch if I find a solution which works for
both lm93 and lineage-pem and is not too complicated.

Guenter


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors