(unknown),

(unknown),

[Andrew Worsley]

  This patch clears the remaining i2c buffer overrun problems that I see in my
hardware.  When run at 200kHz over 2 days and 17 hours there were *NO* faults seen
despite continously accessing the all the i2c devices. I feel the remaining issues
are related to the TPM not behaving properly at clock speeds of 285kHz or higher.
The other i2c hardware is fine up to maximum 400khz.  At these higher clock speeds
the TPM appears to fall behind and I see SDA held low after the TPM read and the
driver report bus arbitration lost errors.  Eventually the TPM completely stops
responding and SDA is held low. But accessing the other i2c hardware causes more
i2c clock pulses which lets the SDA go high again then the other i2c devices work
with out problems which further confirms our thinking that the TPM is source of the
remaining i2c problems.

  With the additional i2c fixes in the attached patch the Xilinx i2c driver
is working with out problems on our hardware. I recommend you consider adding these
changes which apply on top of the previous fixes that I sent.

  Thanks

        Andrew Worsley

[PATCH] Further cadence i2c buffer overrun fixes

[Andrew Worsley]

After this set of changes ran over 2 days 17 hours with no i2c failures
at 200kHz. Previously would lock up with BA (Bus Active) bit set
continously with in a few hours

Set the transfer size register *before* triggering the operation by
setting the i2c address register with the slave address.

If the transfer is less than one FIFO length clear the hold bit before
starting the transfer by writing to the i2c address register.

Finally for transfer larger than a FIFO in size clear the hold bit
just before removing the byte that allows the rest of the transfer
to fit into the FIFO and then complete.
---
 drivers/i2c/busses/i2c-cadence.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/drivers/i2c/busses/i2c-cadence.c b/drivers/i2c/busses/i2c-cadence.c
index 925f77dfded8..260bd6c67e1f 100644
--- a/drivers/i2c/busses/i2c-cadence.c
+++ b/drivers/i2c/busses/i2c-cadence.c
@@ -449,10 +449,10 @@ static irqreturn_t cdns_i2c_master_isr(void *ptr)
 				break;
 			}
 
-			*(id->p_recv_buf)++ =
-				cdns_i2c_readreg(CDNS_I2C_DATA_OFFSET);
 			if (check_hold && id->recv_count == CDNS_I2C_FIFO_DEPTH + 1)
 				cdns_i2c_clear_bus_hold(id);
+			*(id->p_recv_buf)++ =
+				cdns_i2c_readreg(CDNS_I2C_DATA_OFFSET);
 			id->recv_count--;
 			id->curr_recv_count--;
 
@@ -492,10 +492,6 @@ static irqreturn_t cdns_i2c_master_isr(void *ptr)
 		} else if (id->recv_count && !hold_quirk &&
 						!id->curr_recv_count) {
 
-			/* Set the slave address in address register*/
-			cdns_i2c_writereg(id->p_msg->addr & CDNS_I2C_ADDR_MASK,
-						CDNS_I2C_ADDR_OFFSET);
-
 			if (id->recv_count > CDNS_I2C_TRANSFER_SIZE) {
 				cdns_i2c_writereg(CDNS_I2C_TRANSFER_SIZE,
 						CDNS_I2C_XFER_SIZE_OFFSET);
@@ -505,6 +501,10 @@ static irqreturn_t cdns_i2c_master_isr(void *ptr)
 						CDNS_I2C_XFER_SIZE_OFFSET);
 				id->curr_recv_count = id->recv_count;
 			}
+
+			/* write the address register - triggers operation */
+			cdns_i2c_writereg(id->p_msg->addr & CDNS_I2C_ADDR_MASK,
+						CDNS_I2C_ADDR_OFFSET);
 		}
 
 		/* Clear hold (if not repeated start) and signal completion */
@@ -661,15 +661,15 @@ static void cdns_i2c_mrecv(struct cdns_i2c *id)
 		cdns_i2c_writereg(id->recv_count, CDNS_I2C_XFER_SIZE_OFFSET);
 	}
 
-	/* Set the slave address in address register - triggers operation */
-	cdns_i2c_writereg(CDNS_I2C_ENABLED_INTR_MASK, CDNS_I2C_IER_OFFSET);
-	cdns_i2c_writereg(id->p_msg->addr & CDNS_I2C_ADDR_MASK,
-						CDNS_I2C_ADDR_OFFSET);
 	/* Clear the bus hold flag if bytes to receive is less than FIFO size */
 	if (!id->bus_hold_flag &&
 		((id->p_msg->flags & I2C_M_RECV_LEN) != I2C_M_RECV_LEN) &&
 		(id->recv_count <= CDNS_I2C_FIFO_DEPTH))
 			cdns_i2c_clear_bus_hold(id);
+	/* Set the slave address in address register - triggers operation */
+	cdns_i2c_writereg(CDNS_I2C_ENABLED_INTR_MASK, CDNS_I2C_IER_OFFSET);
+	cdns_i2c_writereg(id->p_msg->addr & CDNS_I2C_ADDR_MASK,
+						CDNS_I2C_ADDR_OFFSET);
 }
 
 /**
-- 
2.11.0

Re: [PATCH] Further cadence i2c buffer overrun fixes

[Shubhrajyoti Datta]

Hi Andrew,

On Mon, Apr 16, 2018 at 6:55 AM Andrew Worsley <amworsley@gmail.com> wrote:
>
> After this set of changes ran over 2 days 17 hours with no i2c failures
> at 200kHz.
Can you explain this speed why you need it.
why not the standard ones.

>  Previously would lock up with BA (Bus Active) bit set
> continously with in a few hours
>