i2c: mvtwsi: fix reading status register after interrupt
authorMarek Behún <marek.behun@nic.cz>
Thu, 2 May 2019 14:53:38 +0000 (16:53 +0200)
committerStefan Roese <sr@denx.de>
Fri, 3 May 2019 06:14:39 +0000 (08:14 +0200)
The twsi_wait function reads the control register for interrupt flag,
and if interrupt flag is present, it immediately reads status register.

On our device this sometimes causes bad value being read from status
register, as if the value was not yet updated.

My theory is that the controller does approximately this:
  1. sets interrupt flag in control register,
  2. sets the value of status register,
  3. causes an interrupt

In U-Boot we do not use interrupts, so I think that it is possible that
sometimes the status register in the twsi_wait function is read between
points 1 and 2.

The bug does not appear if I add a small delay before reading status
register.

Wait 100ns (which in U-Boot currently means 1 us, because ndelay(i)
function calls udelay(DIV_ROUND_UP(i, 1000))) before reading the status
register.

Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Heiko Schocher <hs@denx.de>
Reviewed-by: Stefan Roese <sr@denx.de>
Cc: Mario Six <mario.six@gdsys.cc>
Cc: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Stefan Roese <sr@denx.de>
drivers/i2c/mvtwsi.c

index 74ac0a4aa78976487f8314ec7c531a7f781d9299..0a2dafcec6caceb1b0b25f1417112ae8da7657af 100644 (file)
@@ -271,6 +271,17 @@ static int twsi_wait(struct mvtwsi_registers *twsi, int expected_status,
        do {
                control = readl(&twsi->control);
                if (control & MVTWSI_CONTROL_IFLG) {
+                       /*
+                        * On Armada 38x it seems that the controller works as
+                        * if it first set the MVTWSI_CONTROL_IFLAG in the
+                        * control register and only after that it changed the
+                        * status register.
+                        * This sometimes caused weird bugs which only appeared
+                        * on selected I2C speeds and even then only sometimes.
+                        * We therefore add here a simple ndealy(100), which
+                        * seems to fix this weird bug.
+                        */
+                       ndelay(100);
                        status = readl(&twsi->status);
                        if (status == expected_status)
                                return 0;