On 04.06.24 00:37, Andi Shyti wrote:
Hi Gerhard,
Hello Andi
On Sat, Jun 01, 2024 at 09:28:45PM +0200, Gerhard Engleder wrote:
From: Gerhard Engleder <eg@xxxxxxxx>
The KEBA I2C controller is found in the system FPGA of KEBA PLC devices.
It is used to connect EEPROMs and hardware monitoring chips.
can you please add more information about the device, please?
I will add more information
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) KEBA AG 2012
can we update the date here?
First driver version is from 2012. I will remove that line.
+ * Copyright (C) KEBA Industrial Automation Gmbh 2024
+ *
+ * Driver for KEBA I2C controller FPGA IP core
+ */
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/io.h>
+#include <linux/iopoll.h>
+#include <linux/i2c.h>
+#include <linux/platform_data/i2c-keba.h>
Can you sort them in alphabetical order, please?
Will be done.
+#define KI2C "i2c-keba"
+
+#define KI2C_CAPABILITY_REG 0x02
+#define KI2C_CONTROL_REG 0x04
+#define KI2C_CONTROL_DC_REG 0x05
+#define KI2C_STATUS_REG 0x08
+#define KI2C_STATUS_DC_REG 0x09
+#define KI2C_DATA_REG 0x0c
+
+#define KI2C_CAPABILITY_CRYPTO 0x01
This crypto is not used anywhere, did you add it for completness
or have you forgotten to use it?
It is there for completeness/documentation.
+#define KI2C_CAPABILITY_DC 0x02
+
+#define KI2C_CONTROL_MEN 0x01
+#define KI2C_CONTROL_MSTA 0x02
+#define KI2C_CONTROL_RSTA 0x04
+#define KI2C_CONTROL_MTX 0x08
+#define KI2C_CONTROL_TXAK 0x10
+
+#define KI2C_STATUS_IN_USE 0x01
+#define KI2C_STATUS_ACK_CYC 0x02
+#define KI2C_STATUS_RXAK 0x04
+#define KI2C_STATUS_MCF 0x08
+
+#define KI2C_DC_SDA 0x01
+#define KI2C_DC_SCL 0x02
You could eventually make it as:
#define REG1_ADDR 0xXX
#define REG1_VAL_1 0xXX
#define REG1_VAL_2 0xXX
#define REG1_VAL_3 0xXX
#define REG2_ADDR 0xXX
#define REG2_VAL_1 0xXX
#define REG2_VAL_2 0xXX
#define REG2_VAL_3 0xXX
So that it's clear what belongs to what. Not a binding comment,
just an aesthetic note.
I will give it a try.
+
+#define KI2C_INUSE_SLEEP_US (2 * USEC_PER_MSEC)
+#define KI2C_INUSE_TIMEOUT_US (10 * USEC_PER_SEC)
+
+#define KI2C_POLL_DELAY_US 5
+
+struct ki2c {
+ struct platform_device *pdev;
+ void __iomem *base;
+ struct i2c_adapter adapter;
+
+ struct i2c_client **client;
+ int client_size;
+};
+
+static int ki2c_inuse_lock(struct ki2c *ki2c)
+{
+ u8 sts;
+ int ret;
+
+ /* The I2C controller has an IN_USE bit for locking access to the
+ * controller. This enables the use of I2C controller by other none
+ * Linux processors.
Please use the proper commenting style:
/*
* Comment line 1
First driver version is from 2012. I will
* Comment line 2
* ...
* Comment line N
*/
Sorry, I forgot that only net is using that style. Will be changed for
all comments.
+ *
+ * If the I2C controller is free, then the first read returns
+ * IN_USE == 0. After that the I2C controller is locked and further
+ * reads of IN_USE return 1.
+ *
+ * The I2C controller is unlocked by writing 1 into IN_USE.
+ */
Basically this is a semaphore.
I will enhance the comment.
+ ret = readb_poll_timeout(ki2c->base + KI2C_STATUS_REG,
+ sts, (sts & KI2C_STATUS_IN_USE) == 0,
+ KI2C_INUSE_SLEEP_US, KI2C_INUSE_TIMEOUT_US);
we are waiting too long here... the documentaition recommends to
use the readb_poll_timeout for less than 10us, while we are
waiting 2ms.
I will check if it can be changed. Should be possible.
+ if (ret != 0)
+ dev_warn(&ki2c->pdev->dev, "%s err!\n", __func__);
+
+ return ret;
+}
+
+static void ki2c_inuse_unlock(struct ki2c *ki2c)
+{
+ /* unlock the controller by writing 1 into IN_USE */
+ iowrite8(KI2C_STATUS_IN_USE, ki2c->base + KI2C_STATUS_REG);
+}
+
+static int ki2c_wait_for_bit(u8 mask, void __iomem *addr, unsigned long timeout)
It looks more natural to have "addr" as a first argument.
I will reorder.
+{
+ u8 val;
+
+ return readb_poll_timeout(addr, val, (val & mask), KI2C_POLL_DELAY_US,
+ jiffies_to_usecs(timeout));
+}
+static int ki2c_get_sda(struct ki2c *ki2c)
+{
+ /* capability KI2C_CAPABILITY_DC required */
+ return (ioread8(ki2c->base + KI2C_STATUS_DC_REG) & KI2C_DC_SDA) != 0;
Please avoid using such compact style.
Will make it more readable.
+}
+ /* generate clock cycles */
+ ki2c_set_scl(ki2c, val);
+ ndelay(KI2C_RECOVERY_NDELAY);
+ while (count++ < KI2C_RECOVERY_CLK_CNT * 2) {
+ if (val) {
+ /* SCL shouldn't be low here */
+ if (!ki2c_get_scl(ki2c)) {
+ dev_err(&ki2c->pdev->dev,
+ "SCL is stuck low!\n");
+ ret = -EBUSY;
+ break;
+ }
+
+ /* break if SDA is high */
+ if (ki2c_get_sda(ki2c))
+ break;
+ }
+
+ val = !val;
+ ki2c_set_scl(ki2c, val);
+ ndelay(KI2C_RECOVERY_NDELAY);
I don't know how much sense it makes to wait in ndelays, this is
not going to be precise and... are we sure we want to wait
atomically here?
So far there were no problems so it should be precise enough. Delay
is only 5us so sleeping is not necessary. This is done during startup,
sleeping would delay startup.
+ }
+
+ if (!ki2c_get_sda(ki2c)) {
+ dev_err(&ki2c->pdev->dev, "SDA is still low!\n");
To me this and the above dev_err's are just spamming the dmesg as
we are already printing up in the probe function. If we want to
have more precision printing, then we need to chose where the
dev_err's need to be.
I will improve the error reporting.
+ ret = -EBUSY;
+ }
+
+ /* reenable controller */
+ iowrite8(KI2C_CONTROL_MEN, ki2c->base + KI2C_CONTROL_REG);
...
+ ret = ki2c_wait_for_data_ack(ki2c);
+ if (ret < 0)
+ /* For EEPROMs this is normal behavior during internal write
+ * operation.
Please, mind the coding style.
I will do.
+ */
+ dev_dbg(&ki2c->pdev->dev, "%s wait for ACK err at 0x%02x!\n",
+ __func__, m->addr);
+
+ return ret;
+}
...
+static int ki2c_probe(struct platform_device *pdev)
+{
+ struct i2c_keba_platform_data *pdata;
+ struct device *dev = &pdev->dev;
+ struct i2c_adapter *adap;
+ struct resource *io;
+ struct ki2c *ki2c;
+ int ret;
+
+ pdata = dev->platform_data;
+ if (pdata == 0) {
+ dev_err(dev, "Platform data not found!\n");
+ return -ENODEV;
please use dev_err_probe()
This function is new to me. I will check.
Thank you for your review!
Gerhard
