We already set DPM_FLAG_SMART_PREPARE, so we completely skip all
callbacks (other then prepare) where possible, quoting from
dw_i2c_plat_prepare():

        /*
         * If the ACPI companion device object is present for this device, it
         * may be accessed during suspend and resume of other devices via I2C
         * operation regions, so tell the PM core and middle layers to avoid
         * skipping system suspend/resume callbacks for it in that case.
         */
        return !has_acpi_companion(dev);

Also setting the DPM_FLAG_SMART_SUSPEND will cause acpi_subsys_suspend()
to leave the controller runtime-suspended even if dw_i2c_plat_prepare()
returned 0.

Leaving the controller runtime-suspended normally, when the I2C controller
is suspended during the suspend_late phase, is not an issue because
the pm_runtime_get_sync() done by i2c_dw_xfer() will (runtime-)resume it.

But for dw I2C controllers on Bay- and Cherry-Trail devices acpi_lpss.c
leaves the controller alive until the suspend_noirq phase, because it may
be used by the _PS3 ACPI methods of PCI devices and PCI devices are left
powered on until the suspend_noirq phase.

Between the suspend_late and resume_early phases runtime-pm is disabled.
So for any ACPI I2C OPRegion accesses done after the suspend_late phase,
the pm_runtime_get_sync() done by i2c_dw_xfer() is a no-op and the
controller is left runtime-suspended.

i2c_dw_xfer() has a check to catch this condition (rather then waiting
for the I2C transfer to timeout because the controller is suspended).
acpi_subsys_suspend() leaving the controller runtime-suspended in
combination with an ACPI I2C OPRegion access done after the suspend_late
phase triggers this check, leading to the following error being logged
on a Bay Trail based Lenovo Thinkpad 8 tablet:

[   93.275882] i2c_designware 80860F41:00: Transfer while suspended
[   93.275993] WARNING: CPU: 0 PID: 412 at drivers/i2c/busses/i2c-designware-master.c:429 i2c_dw_xfer+0x239/0x280
...
[   93.276252] Workqueue: kacpi_notify acpi_os_execute_deferred
[   93.276267] RIP: 0010:i2c_dw_xfer+0x239/0x280
...
[   93.276340] Call Trace:
[   93.276366]  __i2c_transfer+0x121/0x520
[   93.276379]  i2c_transfer+0x4c/0x100
[   93.276392]  i2c_acpi_space_handler+0x219/0x510
[   93.276408]  ? up+0x40/0x60
[   93.276419]  ? i2c_acpi_notify+0x130/0x130
[   93.276433]  acpi_ev_address_space_dispatch+0x1e1/0x252
...

So since on BYT and CHT platforms we want ACPI I2c OPRegion accesses
to work until the suspend_noirq phase, we need the controller to be
runtime-resumed during the suspend phase if it is runtime-suspended
suspended at that time. This means that we must not set the
DPM_FLAG_SMART_SUSPEND on these platforms.

On BYT and CHT we already have a special ACCESS_NO_IRQ_SUSPEND flag
to make sure the controller stays functional until the suspend_noirq
phase. This commit makes the driver not set the DPM_FLAG_SMART_SUSPEND
flag when that flag is set.

Cc: stable@vger.kernel.org
Fixes: b30f2f65 ("i2c: designware: Set IRQF_NO_SUSPEND flag for all BYT and CHT controllers")
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>