[PATCH] i2c: designware: Don't mask TX_EMPTY if write is in progress

[Jisheng Zhang]

Dear Jarkko,

On Fri, 30 Oct 2015 14:03:53 +0200
Jarkko Nikula wrote:

> Hi
> 
> On 10/30/2015 09:37 AM, Jisheng Zhang wrote:
> > Currently when i2c_msg index search is completed, TX_EMPTY interrupt
> > will be masked. But if the size of i2c_msg data is longer than the size
> > of the tx buffer, we still need TX_EMPTY interrupt, otherwise we will
> > get "controller timed out" error.
> >
> > This patch fixes this issue by only masking TX_EMPTY if i2c_msg index
> > search is completed and write is not in progress.
> >
> > Signed-off-by: Jisheng Zhang <jszhang at marvell.com>
> > ---
> >   drivers/i2c/busses/i2c-designware-core.c | 7 ++++---
> >   1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/i2c/busses/i2c-designware-core.c b/drivers/i2c/busses/i2c-designware-core.c
> > index 7441cdc..a2eb212 100644
> > --- a/drivers/i2c/busses/i2c-designware-core.c
> > +++ b/drivers/i2c/busses/i2c-designware-core.c
> > @@ -542,10 +542,11 @@ i2c_dw_xfer_msg(struct dw_i2c_dev *dev)
> >   	}
> >
> >   	/*
> > -	 * If i2c_msg index search is completed, we don't need TX_EMPTY
> > -	 * interrupt any more.
> > +	 * If i2c_msg index search is completed and writing is not in progress,
> > +	 * we don't need TX_EMPTY interrupt any more.
> >   	 */
> > -	if (dev->msg_write_idx == dev->msgs_num)
> > +	if (dev->msg_write_idx == dev->msgs_num &&
> > +			!(dev->status & STATUS_WRITE_IN_PROGRESS))
> >   		intr_mask &= ~DW_IC_INTR_TX_EMPTY;
> >  
> This looks a possible scenario here. What I'm wondering how is the 
> transfer ending and what condition in i2c_dw_isr() causes the complete() 
> call?
> 
> Assuming we break the for loop in i2c_dw_xfer_msg() with buf_len > 0 and 
> dev->msg_write_idx == dev->msgs_num how we get back inside the for loop 
> for writing the remaining bytes and stop command?

This is a good question. After more consideration, I think the following
condition can't exist:

  (msg_write_idx == dev->msgs_num) && (dev->status & STATUS_WRITE_IN_PROGRESS)

so this patch isn't necessary.

Thanks for your review,
Jisheng