On Sat, 5 Apr 2014 09:13:16 +0300
"Westerberg, Mika" <mika.westerberg@intel.com> wrote:

> On Sat, Apr 05, 2014 at 12:54:33AM +0300, Du, Wenkai wrote:
> > >Interrupt masking is done already after each transaction.
> > 
> > At end of transfer, the code uses __i2c_dw_enable(dev, false) to disable
> > adapter. This function doesn't mask interrupts. There is another function
> > i2c_dw_disable that masks and clears interrupts. This could be used, but
> > that means we need to fix in 2 places: 
> 
> Please check i2c_dw_isr() and tell me in which code path interrupts are not
> getting masked. Or am I missing something fundamental here?
> 
> In case of abort, we mask interrupts. Also whenever the transaction
> completes we mask interrupts (in i2c_dw_xfer_msg()).

Well actually you mask the IRQ at some point after the function returns
if the bus allows the write to be posted. As i2c_dw_isr can then exit the
IRQ handler before the write completes I suspect you have a race ?

Alan