[Gumstix-users] Bluetooth fix for OE 318, kernel 2.6.24 on Verdex

[<pa...@pa...>]

Hi,

After upgrading OE rev 318 from 2.6.21 to 2.6.24 on a Verdex
I got the infamous "No response from BT module".

The problem is that the pxa serial driver fails to enable the
BTUART clock, as seen here:

|   $ cat /dev/ttyS1 &   # Keeps the BTUART open.
|   $ pxaregs CKEN_7
|   CM BTUART clock enabled
|   CKEN_7                   0x00500640  00000000 01010000 00000110 01000000
|   CKEN_7                            0  CM BTUART clock enabled # Should be 1

As a workaround, run "pxaregs CKEN_7 1" before starting hciattach.


This is caused by a refactoring of clocking code in recent kernels
(presumably in order to improve power management):

linux-2.6.24/arch/arm/mach-pxa/pxa27x.c
|   static struct clk pxa27x_clks[] = {
|           [...]
|           INIT_CKEN("UARTCLK", FFUART, 14857000, 1, &pxa_device_ffuart.dev),
|           INIT_CKEN("UARTCLK", BTUART, 14857000, 1, &pxa_device_btuart.dev),
|           INIT_CKEN("UARTCLK", STUART, 14857000, 1, NULL),

Note that UART clocks are distinguished only by their platform_device,
except STUART which serves as a wildcard. Unfortunately the array is
transformed into a reversed list, and as a result clk_get() always
matches STUART first.

Apparently this is/was a problem on PXA255 too:
http://www.spinics.net/lists/arm-kernel/msg47446.html

Things have changed again in the latest GIT kernel, so hopefully
this is already fixed, but I anyone with the "No response from
BT module" should remember to check CKEN_7.

Pascal