Re: I2c-ocore linux driver debug

[Peter Korsgaard <peter.korsgaard@barco.com>]

>>>>> "FM" == Fabien Marteau <lepingouin.tux@gmail.com> writes:

Hi,

Discussions about the Linux i2c drivers belongs on the
linux-i2c@vger.kernel.org list, let's move discussion there.

FM> I integrated the IP i2c ocore for my platform ARMadeus apf9328
FM> (http://www.armadeus.com/wiki).  ARMadeus platform is an embedded
FM> system with a processor i.MXL and a FPGA Spartan3.

FM> The interface between the processor and FPGA is a 16bits data bus,
FM> and it's wired to do only 16bits read/write. Thus, I had to write
FM> a wrapper to convert 8bits ocore interface in 16bits
FM> interface. This wrapper mutiplex data bus to keep data in low part
FM> of the 16bits bus (see attachement) and add an ID register to
FM> unsure driver that the component is configured correctly.

FM> I had two problems with the driver:
FM> * Registers addresses.
FM> * To send stop command, driver have to set stop bit *before*
FM> sending the last byte. In the driver, this stop bit is set at the
FM> end of command, and that doesn't work.

FM> I was really surprised of this bugs, because this IP is old and I
FM> feel I am the only one that have these problems. Maybe the problem
FM> come from wrapper but I don't know why.

FM> I modified the driver to fix these bugs (see diff in attachment), and that
FM> work well with two components (LIS3LV02DL accelerometer and DS28CZ04 EEPROM)
FM> and testbench provided in attachment.

FM> Addresses registers problem
FM> ===========================

FM> The original driver give these addresses:
FM> /* registers */
FM> #define OCI2C_PRELOW		0
FM> #define OCI2C_PREHIGH		1
FM> #define OCI2C_CONTROL		2
FM> #define OCI2C_DATA		3
FM> #define OCI2C_CMD		4 /* write only */
FM> #define OCI2C_STATUS		4 /* read only, same address as OCI2C_CMD */

That fits the specs (i2c_specs.pdf, by Richard Herveille)

FM> After lot of tests, I found that this addresses work better:
FM> /* registers */
FM> #define OCI2C_PRELOW		0
FM> #define OCI2C_PREHIGH		1
FM> #define OCI2C_CONTROL		2
FM> #define OCI2C_CMD			4
FM> #define OCI2C_DATA_TXR		5
FM> #define OCI2C_STATUS		6
FM> #define OCI2C_DATA_RXR		7

Strange, seems like something went wrong in your wrapper.

FM> Stop command
FM> ============

FM> The most important problem I had was the stop command. To send a command like
FM> write quick SMBus (see Documentation/i2c/smbus-protocol) :

FM> S Addr [Rd/Wr] P

FM> The stop bit must be set at the same time that start bit. To do
FM> that I modified
FM> the xfer function:
FM> 	oc_setreg(i2c, OCI2C_CMD, OCI2C_CMD_START);
FM> by:
FM> 	if(i2c->msg->len == 0)
FM> 		oc_setreg(i2c, OCI2C_CMD, OCI2C_CMD_START|OCI2C_CMD_STOP);
FM> 	else
FM> 		oc_setreg(i2c, OCI2C_CMD, OCI2C_CMD_START);

It could be that we have a problem with zero length transfers (don't
have access to any smbus devices) - Could you please send a proper
patch in diff -urpN format?

-- 
Bye, Peter Korsgaard