Re: [PATCH V9 4/5] i2c: tegra: update transfer timeout

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1458 bytes --]

On Fri, Feb 01, 2019 at 07:37:45PM +0000, Sowjanya Komatineni wrote:
> 
> > > BYTES_PER_FIFO_WORD 4
> > >  
> > >  #define I2C_CNFG				0x000
> > > @@ -893,8 +892,9 @@ static int tegra_i2c_issue_bus_clear(struct tegra_i2c_dev *i2c_dev)
> > >  		i2c_writel(i2c_dev, reg, I2C_BUS_CLEAR_CNFG);
> > >  		tegra_i2c_unmask_irq(i2c_dev, I2C_INT_BUS_CLR_DONE);
> > >  
> > > -		time_left = wait_for_completion_timeout(&i2c_dev->msg_complete,
> > > -							TEGRA_I2C_TIMEOUT);
> > > +		time_left = wait_for_completion_timeout(
> > > +						&i2c_dev->msg_complete,
> > > +						msecs_to_jiffies(1000));
> >
> > So potentially tegra_i2c_xfer_msg() could take more than 1 second
> > and then fail with -EAGAIN, correct? In that case we should set
> > adapter.timeout in probe to a larger value:
> >
> Bus clear pulse threshold we are setting is 9 (default as per spec) so
> bus clear should happen after finishing sending of 9 pulses.
> So 1sec is very long time for bus to get released and in case of ARB
> LOST whole transfer to return EAGAIN will be much less then 1sec

What if for example we have a very long transfer, say some 64 KiB on
Tegra194 and about 95% through the transfer something causes the bus
to lock up. The transfer so far would've taken something on the order
of 5 seconds, but we'd still fail, potentially with -EAGAIN, right?

Or is there some other mechanism that would prevent the above from
happening?

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]