Re: [RFC: 2.6 patch] remove the i2c-at91.c driver

[David Brownell <david-b-yBeKhBN/0LDR7s880joybQ@public.gmane.org>]

On Thursday 04 September 2008, Andrew Victor wrote:
> 
> I don't actually have G20 hardware to test it on, but until
> somebody/Atmel have verified if this driver works or not, I'd prefer
> to just to leave it as broken.
> If it still doesn't work, then just remove all references to the driver.

Right ...

Note that resolving the lack of support for repeated starts
would require driver updates in any case.  Some of which COULD
have been done with existing hardware.  ISTR a combined
transaction could be done

	S addr W [A] byte1 [A] byte2 [A] byte3 [A]
	Sr addr R [A] [byte4] A [byte5] ... N P

... with up to three bytes in the write segment.  Obviously
that's not an arbitrary I2C transaction ... in fact, it'd
best be handled (IMO) as an SMBus adapter not an I2C adapter.
(So maybe a new driver would be appropriate...)

But I recall the lack of flow control to be a bigger issue.
The hardware liked to TX the next byte before it was loaded
into the (single byte) TX fifo, and then complain that there
was no byte there.  And likewise, to RX the next byte before
the previous one got removed.  Or ... if that's not what it
was doing, it sure acted like it was!  Now, *THOSE* issues
seem highly amenable to hardware-only fixes:  just stretch
the clock until the FIFO is ready.

- Dave

_______________________________________________
i2c mailing list
i2c-GZX6beZjE8VD60Wz+7aTrA@public.gmane.org
http://lists.lm-sensors.org/mailman/listinfo/i2c