Re: [PATCH V1] i2c: busses: tegra: Add suspend-resume support

[Bitan Biswas]

On 5/31/19 5:43 AM, Dmitry Osipenko wrote:
> 31.05.2019 11:50, Bitan Biswas ÐÐÑÐÑ:
> > On 5/30/19 4:27 AM, Dmitry Osipenko wrote:
> > > 30.05.2019 8:55, Bitan Biswas ÐÐÑÐÑ:
> > > > Post suspend I2C registers have power on reset values. Before any
> > > > transfer initialize I2C registers to prevent I2C transfer timeout
> > > > and implement suspend and resume callbacks needed. Fix below errors
> > > > post suspend:
> > > >
> > > > 1) Tegra I2C transfer timeout during jetson tx2 resume:
> > > >
> > > > [ÂÂ 27.520613] pca953x 1-0074: calling pca953x_resume+0x0/0x1b0 @
> > > > 2939, parent: i2c-1
> > > > [ÂÂ 27.633623] tegra-i2c 3160000.i2c: i2c transfer timed out
> > > > [ÂÂ 27.639162] pca953x 1-0074: Unable to sync registers 0x3-0x5. -110
> > > > [ÂÂ 27.645336] pca953x 1-0074: Failed to sync GPIO dir registers: -110
> > > > [ÂÂ 27.651596] PM: dpm_run_callback(): pca953x_resume+0x0/0x1b0
> > > > returns -110
> > > > [ÂÂ 27.658375] pca953x 1-0074: pca953x_resume+0x0/0x1b0 returned -110
> > > > after 127152 usecs
> > > > [ÂÂ 27.666194] PM: Device 1-0074 failed to resume: error -110
> > > >
> > > > 2) Tegra I2C transfer timeout error on jetson Xavier post resume.
> > > >
> > > > Signed-off-by: Bitan Biswas <bbiswas@xxxxxxxxxx>
> > > > ---
> > > >  Â drivers/i2c/busses/i2c-tegra.c | 24 ++++++++++++++++++++++++
> > > >  Â 1 file changed, 24 insertions(+)
> > > >
> > > > diff --git a/drivers/i2c/busses/i2c-tegra.c
> > > > b/drivers/i2c/busses/i2c-tegra.c
> > > > index ebaa78d..f6a377f 100644
> > > > --- a/drivers/i2c/busses/i2c-tegra.c
> > > > +++ b/drivers/i2c/busses/i2c-tegra.c
> > > > @@ -1687,9 +1687,33 @@ static int tegra_i2c_remove(struct
> > > > platform_device *pdev)
> > > >  Â }
> > > >  Â Â #ifdef CONFIG_PM_SLEEP
> > > > +static int tegra_i2c_suspend(struct device *dev)
> > > > +{
> > > > +ÂÂÂ struct tegra_i2c_dev *i2c_dev = dev_get_drvdata(dev);
> > > > +
> > > > +ÂÂÂ i2c_mark_adapter_suspended(&i2c_dev->adapter);
> > > > +
> > > > +ÂÂÂ return 0;
> > > > +}
> > > > +
> > > > +static int tegra_i2c_resume(struct device *dev)
> > > > +{
> > > > +ÂÂÂ struct tegra_i2c_dev *i2c_dev = dev_get_drvdata(dev);
> > > > +ÂÂÂ int ret;
> > > > +
> > > > +ÂÂÂ i2c_lock_bus(&i2c_dev->adapter, I2C_LOCK_ROOT_ADAPTER);
> > > > +ÂÂÂ ret = tegra_i2c_init(i2c_dev, false);
> > > > +ÂÂÂ i2c_unlock_bus(&i2c_dev->adapter, I2C_LOCK_ROOT_ADAPTER);
> > >
> > > Why the locking is needed here?
> >
> > async resume could result in stress test issues if some client accesses
> > the i2c instance. This ensures the i2c instance is locked till the
> > initialization is complete.
>
> 1) This doesn't make much sense.. if client could access I2C during of
> tegra_i2c_init execution, then what stops it to perform the access
> before the lock is taken?
Client resumes will start after I2C instance resume because of driver 
dependency. Since lock is the first call in i2c-tegra I believe I2C 
calls of client will not start.

> 2) The whole point of the i2c_mark_adapter_* API is to catch those
> faulty clients that have a broken suspend-resume sequence. Client will
> get a -ESHUTDOWN on trying to issue I2C transfer while controller is
> marked as suspended.
i2c lock bus calls were used in the resume callback implementation that 
was reverted few months back. Hence, these were added in this patch 
which should be more like a revert-of-revert .

But I feel probably your point also makes sense. Old resume callback did 
not have i2c_mark_adapter_* calls. Based on the i2c_adapter_mark_* API 
documentation it should be taking care that core i2c calls from client 
are not started. I plan to update the patch and remove the lock-unlock 
guards in resume callback.


> 3) Please don't use async suspend-resume where it doesn't make sense.
This is a system wide setting. /sys/power/pm_async by default is 1 and 
there is no driver specific change in this patch to choose async 
suspend-resume.

> Corollary: you should drop the locking because it doesn't do anything
> useful.
I did some basic suspend resume tests and do not see any problems 
removing the i2c_lock_bus call you pointed out.

-Thanks,
 Bitan