Re: ALL THE I2C BUSSES

[Jason Thorpe <thorpej%me.com@localhost>]

> On Apr 9, 2018, at 7:48 PM, Jason Thorpe <thorpej%me.com@localhost> wrote:
> 
> Forgive me, folks, I’m a little new to the Raspberry Pi (a little slow on the uptake, I guess :-)…
> 
> I’m a little confused about how to map the prevailing documentation on the RPI and it’s Broadcom SoC to NetBSD’s treatment of the device tree.  NetBSD shows 3 i2c controllers (iic0, iic1, iic2).  I also know the 40-pin Pi models with newer firmware have a dedicated ID EERPOM i2c bus brought out to pins 27 and 28.  All of the pinout diagrams I’ve seen show the i2c that’s on pins 3 and 5 as SDA1 / SCL1.  I always assumed that the first i2c bus wasn’t brought out to the 40-pin header, but rather used on-board only.
> 
> So, I figured that the i2c on the header would be “iic1”, reinforced by the RPI kernel config file’s dsrtc* entry.
> 
> Well, much to my chagrin, dsrtc0 was not, in fact, found on iic1.  i2cscan found them on iic2.  I can fix the kernel config file, no big deal.  But I’d like to understand the device unit mapping at play.

There’s also a bug.  The “iic” driver supports both direct configuration of devices (for board configs with a known set of devices on the i2c), as well as indirect config, which searches for devices specified in the kernel config file.

It looks like this code has changed a bit since I wrote it 15 years ago.  If the parent driver (bsciic, in this case) provides a prop_array_t as iba->iba_child_devices, then indirect configuration is bypassed.  If it does not, then there’s a dance to check for i2c-indirect-config == false or if the parent has i2c-child-devices as a property on itself, in which case, again, indirect config is skipped.

The problem in the bsciic driver is that if there is not i2c-child-devices property, then an empty array is assigned to iba->iba_child_devices.  This makes the check against NULL fail, and thus no indirect config devices will be found.

The following patch makes the dsrtc driver work correctly on my RPI:

Index: bcm2835_bsc.c
===================================================================
RCS file: /cvsroot/src/sys/arch/arm/broadcom/bcm2835_bsc.c,v
retrieving revision 1.10
diff -u -p -r1.10 bcm2835_bsc.c
--- bcm2835_bsc.c	3 Mar 2018 16:03:38 -0000	1.10
+++ bcm2835_bsc.c	10 Apr 2018 04:22:17 -0000
@@ -184,7 +184,7 @@ bsciic_attach(device_t parent, device_t 
 	if (iba.iba_child_devices)
 		prop_object_retain(iba.iba_child_devices);
 	else
-		iba.iba_child_devices = prop_array_create();
+		iba.iba_child_devices = NULL;
 	prop_object_release(devs);
 
 	config_found_ia(self, "i2cbus", &iba, iicbus_print);

A quick perusal indicates that other board packages have the same bug.

.
.
.
bsciic2 at fdt1: Broadcom Serial Controller
iic2 at bsciic2: I2C bus
dsrtc0 at iic2 addr 0x68: DS3231 Real-time Clock
.
.
.

(and I’ve confirmed that the correct date/time are stored in the RTC at reboot, and read correctly at boot time.)

-- thorpej