Received-SPF: pass (google.com: domain of linux-kernel+bounces-108449-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <65fa759b.5d0a0220.fe5f7.1fa0@mx.google.com>
From: Sam Edwards <cfsworks@gmail.com>
To: Gregory CLEMENT <gregory.clement@bootlin.com>, Andi Shyti <andi.shyti@kernel.org>
Cc: linux-i2c@vger.kernel.org, linux-kernel@vger.kernel.org
Date: Tue, 19 Mar 2024 16:51:51 -0600
Subject: [RESEND v2 RFC 1/5] i2c: mv64xxx: Clear bus errors before transfer
Precedence: bulk

The MV64XXX hardware can act as either a bus controller or target
device. In order to protect target devices from a glitching bus
(apparently), the core listens on the lines even when idle and moves the
hardware FSM to the "BUS_ERR" state if an invalid transition is
detected. The hardware then does not exit this state until reset.

This feature is actually counterproductive when using the hardware as a
controller (as this driver does): we do not actually *care* what
happened on the bus previously, as long as it's ready for use when the
new transfer starts. However, the controller will remember a previous
glitch and trip up the driver after it attempts to send the start
command. The driver logs and error and resets the controller, recovering
from the BUS_ERR state, but not without erroring back the transfer with
errno EAGAIN. Clients generally do not handle this gracefully.

This is easily fixed by checking for the BUS_ERR condition upfront and
issuing the hardware reset before beginning the transfer. This patch
does NOT also call i2c_recover_bus(): the assumption is that the bus is
fine, just the hardware is upset; if the bus is also in a bad state,
this should not pass silently.

Signed-off-by: Sam Edwards <sam@turingpi.com>
---
 drivers/i2c/busses/i2c-mv64xxx.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
index fd8403b07fa6..cfc16909fba3 100644
--- a/drivers/i2c/busses/i2c-mv64xxx.c
+++ b/drivers/i2c/busses/i2c-mv64xxx.c
@@ -753,6 +753,7 @@ mv64xxx_i2c_xfer_core(struct i2c_adapter *adap, struct i2c_msg msgs[], int num)
 {
 	struct mv64xxx_i2c_data *drv_data = i2c_get_adapdata(adap);
 	int rc, ret = num;
+	u32 status;
 
 	rc = pm_runtime_resume_and_get(&adap->dev);
 	if (rc)
@@ -762,6 +763,11 @@ mv64xxx_i2c_xfer_core(struct i2c_adapter *adap, struct i2c_msg msgs[], int num)
 	drv_data->msgs = msgs;
 	drv_data->num_msgs = num;
 
+	/* Calm down the hardware if it was upset by a bus glitch while idle */
+	status = readl(drv_data->reg_base + drv_data->reg_offsets.status);
+	if (status == MV64XXX_I2C_STATUS_BUS_ERR)
+		mv64xxx_i2c_hw_init(drv_data);
+
 	if (mv64xxx_i2c_can_offload(drv_data) && !drv_data->atomic)
 		rc = mv64xxx_i2c_offload_xfer(drv_data);
 	else
-- 
2.43.2