On Mon, Sep 09, 2013 at 12:44:49PM -0600, Stephen Warren wrote:
From: Stephen Warren swarren@nvidia.com
If the WM8903 IRQ is triggered after the device is suspended, or before it is resumed, the underlying I2C bus that the WM8903 is attached to may not be available. This may cause the IRQ handler to emit the following error:
wm8903 0-001a: Failed to read IRQ status: -16... or perhaps even cause a failure inside the I2C driver.
To prevent this, explicitly mask all interrupt sources when entering BIAS_OFF. Also, modify the IRQ handler to immediately return if the device is in BIAS_OFF, so that shared interrupt lines don't cause a similar problem. This requires that WM8903 interrupts be disabled to avoid interrupt storms.
During resume (actually, the transition back to BIAS_STANDBY), restore the desired IRQ mask.
This should be being handled in at a framework level, there's a generic problem here that affects essentially every device out there which might need interrupts over suspend (and causes relatively frequent problems). Your fix also won't help if someone has arranged for the wm8903 mic detect to be a wake source which you normally want in phone or tablet type applications but is where this sort of issue most frequently bites since we'll never turn the device off the interrupt does need to be asserted to kick off the wake.
Drivers usually work around this by disabling their interrupt with disable_irq() between suspend() and suspend_late(), then doing the same thing again between resume_noirq() and resume() covering both the windows where this can happen. This preserves the ability to wake from the interrupt but it's not awesome; ideally genirq would be able to understand that it can't deliver interrupts to a suspended device but then of course at the minute genirq has no idea that a device is even associated with the interrupt.