i2c: rk3x: Handle a spurious start completion interrupt flag
authorOndrej Jirman <megous@megous.com>
Fri, 24 Sep 2021 11:15:27 +0000 (13:15 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 17 Dec 2021 09:14:40 +0000 (10:14 +0100)
commitbca6af4325d6f1d5af95b24eed49bd4fb84c8f61
tree84e5699474f778bf641196b76896f56cbb8e0696
parentd6edec8a7b55871db6e2faf7bd39a2e87b020b2b
i2c: rk3x: Handle a spurious start completion interrupt flag

[ Upstream commit 02fe0fbd8a21e183687925c3a266ae27dda9840f ]

In a typical read transfer, start completion flag is being set after
read finishes (notice ipd bit 4 being set):

trasnfer poll=0
i2c start
rk3x-i2c fdd40000.i2c: IRQ: state 1, ipd: 10
i2c read
rk3x-i2c fdd40000.i2c: IRQ: state 2, ipd: 1b
i2c stop
rk3x-i2c fdd40000.i2c: IRQ: state 4, ipd: 33

This causes I2C transfer being aborted in polled mode from a stop completion
handler:

trasnfer poll=1
i2c start
rk3x-i2c fdd40000.i2c: IRQ: state 1, ipd: 10
i2c read
rk3x-i2c fdd40000.i2c: IRQ: state 2, ipd: 0
rk3x-i2c fdd40000.i2c: IRQ: state 2, ipd: 1b
i2c stop
rk3x-i2c fdd40000.i2c: IRQ: state 4, ipd: 13
i2c stop
rk3x-i2c fdd40000.i2c: unexpected irq in STOP: 0x10

Clearing the START flag after read fixes the issue without any obvious
side effects.

This issue was dicovered on RK3566 when adding support for powering
off the RK817 PMIC.

Signed-off-by: Ondrej Jirman <megous@megous.com>
Reviewed-by: John Keeping <john@metanate.com>
Signed-off-by: Wolfram Sang <wsa@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/i2c/busses/i2c-rk3x.c