Re: Nokia N900 - audio TPA6130A2 problems

[Sebastian Reichel <sre@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1425 bytes --]

Hi,

On Wed, Mar 16, 2016 at 02:33:19PM +0100, Pali Rohár wrote:
> Hi! We found out that tpa6130a2 device is being initialized before
> i2c_2 bus is initialized. So that is reason why tpa6130a2 fails...

What do you mean by initialize? A call to tpa6130a2_probe()? In that
case I wonder about client->adapter. Is it NULL?

> Any idea why kernel first try to initialize one i2c device even before
> bus itself is initialized?

Just dump the stack during the tpa6130a2 initialization and you can
see it in the kernel log:

dump_stack();

---------------

I just had another look at the driver and I think there is a race
condition for tpa6130a2_add_controls() and tpa6130a2_stereo_enable().

As far as I can see both functions check for "tpa6130a2_client !=
NULL". tpa6130a2_client is set before the probe function has finished,
though. Simplified probe:

...
tpa6130a2_client = client;
set_default_regs();
acquire_power_gpio();
acquire_regulator();
tpa6130a2_power(1); // needs tpa6130a2_client
check_device();
tpa6130a2_power(0); // needs tpa6130a2_client
...

If tpa6130a2_add_controls() or tpa6130a2_stereo_enable() is called
after tpa6130a2 probe has started, but before probe has completed,
tpa6130a2_client is set, but not yet initialized. The race condition
can be fixed easily by moving the tpa6130a2_client assignment directly
after the regulator acquisition.

-- Sebastian

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]