On 18. 10. 23, 23:11, Frank Li wrote:
Add new driver to allow tty over i3c master.
What is it good for? Why we should consider this for inclusion at all?
Signed-off-by: Frank Li <Frank.Li@xxxxxxx>
---
Notes:
This patch depend on
https://lore.kernel.org/imx/20231018205929.3435110-1-Frank.Li@xxxxxxx/T/#t
drivers/tty/Kconfig | 8 +
drivers/tty/Makefile | 1 +
drivers/tty/i3c_tty.c | 466 ++++++++++++++++++++++++++++++++++++++++++
3 files changed, 475 insertions(+)
create mode 100644 drivers/tty/i3c_tty.c
diff --git a/drivers/tty/Kconfig b/drivers/tty/Kconfig
index 5646dc6242cd9..6d91fe6a211a1 100644
--- a/drivers/tty/Kconfig
+++ b/drivers/tty/Kconfig
@@ -412,6 +412,14 @@ config RPMSG_TTY
To compile this driver as a module, choose M here: the module will be
called rpmsg_tty.
+config I3C_TTY
+ tristate "tty over i3c"
"TTY over I3C"
+ depends on I3C
+ help
+ Select this options if you'd like use tty over I3C master controller
TTY and add a period.
--- /dev/null
+++ b/drivers/tty/i3c_tty.c
@@ -0,0 +1,466 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2023 NXP.
+ *
+ * Author: Frank Li <Frank.Li@xxxxxxx>
+ */
+
+#include <linux/delay.h>
+#include <linux/kernel.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/i3c/device.h>
+#include <linux/i3c/master.h>
+#include <linux/slab.h>
+#include <linux/console.h>
+#include <linux/serial_core.h>
+#include <linux/interrupt.h>
+#include <linux/workqueue.h>
+#include <linux/tty_flip.h>
+
+static DEFINE_IDR(i3c_tty_minors);
+static DEFINE_MUTEX(i3c_tty_minors_lock);
+
+static struct tty_driver *i3c_tty_driver;
+
+#define I3C_TTY_MINORS 256
+#define I3C_TTY_TRANS_SIZE 16
+#define I3C_TTY_RX_STOP BIT(0)
+#define I3C_TTY_RETRY 20
+#define I3C_TTY_YIELD_US 100
+
+struct ttyi3c_port {
+ struct tty_port port;
+ int minor;
+ unsigned int txfifo_size;
+ unsigned int rxfifo_size;
+ struct circ_buf xmit;
Why can't you use port.xmit_fifo?
+ spinlock_t xlock; /* protect xmit */
+ void *buffer;
+ struct i3c_device *i3cdev;
+ struct work_struct txwork;
+ struct work_struct rxwork;
+ struct completion txcomplete;
+ struct workqueue_struct *workqueue;
Why do you need a special wq? And even, why per port?
+ atomic_t status;
+};
+
+static const struct i3c_device_id i3c_ids[] = {
+ I3C_DEVICE(0x011B, 0x1000, NULL),
+ { /* sentinel */ },
+};
+
+static int i3c_port_activate(struct tty_port *port, struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport = container_of(port, struct ttyi3c_port, port);
+
+ atomic_set(&sport->status, 0);
+
+ return i3c_device_enable_ibi(sport->i3cdev);
+}
+
+static void i3c_port_shutdown(struct tty_port *port)
+{
+ struct ttyi3c_port *sport =
+ container_of(port, struct ttyi3c_port, port);
+
+ i3c_device_disable_ibi(sport->i3cdev);
+}
+
+static void i3c_port_destruct(struct tty_port *port)
+{
+ struct ttyi3c_port *sport =
+ container_of(port, struct ttyi3c_port, port);
+
+ mutex_lock(&i3c_tty_minors_lock);
+ idr_remove(&i3c_tty_minors, sport->minor);
+ mutex_unlock(&i3c_tty_minors_lock);
+}
+
+static const struct tty_port_operations i3c_port_ops = {
+ .shutdown = i3c_port_shutdown,
+ .activate = i3c_port_activate,
+ .destruct = i3c_port_destruct,
+};
+
+static struct ttyi3c_port *i3c_get_by_minor(unsigned int minor)
+{
+ struct ttyi3c_port *sport;
+
+ mutex_lock(&i3c_tty_minors_lock);
+ sport = idr_find(&i3c_tty_minors, minor);
+ mutex_unlock(&i3c_tty_minors_lock);
+
+ return sport;
+}
+
+static int i3c_install(struct tty_driver *driver, struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport;
+ int ret;
+
+ sport = i3c_get_by_minor(tty->index);
+ if (!sport)
+ return -ENODEV;
+
+ ret = tty_standard_install(driver, tty);
+ if (ret)
+ return ret;
+
+ tty->driver_data = sport;
+
+ return 0;
+}
You don't need this at all. (Watch for XXX marks below.)
+static ssize_t i3c_write(struct tty_struct *tty, const unsigned char *buf, size_t count)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+ struct circ_buf *circ = &sport->xmit;
+ unsigned long flags;
+ int c, ret = 0;
+
+ spin_lock_irqsave(&sport->xlock, flags);
+ while (1) {
+ c = CIRC_SPACE_TO_END(circ->head, circ->tail, UART_XMIT_SIZE);
+ if (count < c)
+ c = count;
+ if (c <= 0)
+ break;
+
+ memcpy(circ->buf + circ->head, buf, c);
+ circ->head = (circ->head + c) & (UART_XMIT_SIZE - 1);
+ buf += c;
+ count -= c;
+ ret += c;
+ }
+ spin_unlock_irqrestore(&sport->xlock, flags);
With kfifo, all this would be one line, right?
+
+ if (circ->head != circ->tail)
+ queue_work(sport->workqueue, &sport->txwork);
+
+ return ret;
+}
+
+static int i3c_put_char(struct tty_struct *tty, unsigned char ch)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+ struct circ_buf *circ = &sport->xmit;
+ unsigned long flags;
+ int ret = 0;
+
+ spin_lock_irqsave(&sport->xlock, flags);
+
+ if (sport && CIRC_SPACE(circ->head, circ->tail, UART_XMIT_SIZE) != 0) {
+ circ->buf[circ->head] = ch;
+ circ->head = (circ->head + 1) & (UART_XMIT_SIZE - 1);
+ ret = 1;
+ }
+
+ spin_unlock_irqrestore(&sport->xlock, flags);
+
+ return ret;
+}
+
+static void i3c_flush_chars(struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+
+ queue_work(sport->workqueue, &sport->txwork);
+}
+
+static unsigned int i3c_write_room(struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+ struct circ_buf *circ = &sport->xmit;
+ unsigned long flags;
+ int ret = 0;
+
+ spin_lock_irqsave(&sport->xlock, flags);
+ ret = CIRC_SPACE(circ->head, circ->tail, UART_XMIT_SIZE);
+ spin_unlock_irqrestore(&sport->xlock, flags);
+
+ return ret;
+}
+
+static void i3c_throttle(struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+
+ atomic_or(I3C_TTY_RX_STOP, &sport->status);
+}
+
+static void i3c_unthrottle(struct tty_struct *tty)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+
+ atomic_andnot(I3C_TTY_RX_STOP, &sport->status);
+
+ queue_work(sport->workqueue, &sport->rxwork);
+}
+
+static int i3c_open(struct tty_struct *tty, struct file *filp)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
XXX Here, you can simply do:
struct ttyi3c_port *sport = container_of(tty->port, struct ttyi3c_port,
port);
tty->driver_data = sport;
+ return tty_port_open(&sport->port, tty, filp);
+}
+
+static void i3c_close(struct tty_struct *tty, struct file *filp)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+
+ if (!sport)
+ return;
How can that happen?
+ tty_port_close(tty->port, tty, filp);
+}
+
+static void i3c_wait_until_sent(struct tty_struct *tty, int timeout)
+{
+ struct ttyi3c_port *sport = tty->driver_data;
+
+ wait_for_completion_timeout(&sport->txcomplete, timeout);
+ reinit_completion(&sport->txcomplete);
Does this work in parallel?
+}
+
+static const struct tty_operations i3c_tty_ops = {
+ .install = i3c_install,
+ .open = i3c_open,
+ .close = i3c_close,
+ .write = i3c_write,
+ .put_char = i3c_put_char,
+ .flush_chars = i3c_flush_chars,
+ .write_room = i3c_write_room,
+ .throttle = i3c_throttle,
+ .unthrottle = i3c_unthrottle,
+ .wait_until_sent = i3c_wait_until_sent,
+};
+
+static void i3c_controller_irq_handler(struct i3c_device *dev,
+ const struct i3c_ibi_payload *payload)
+{
+ struct ttyi3c_port *sport = dev_get_drvdata(&dev->dev);
+
+ queue_work(sport->workqueue, &sport->rxwork);
Doesn't ibi provide threaded irqs? Oh, wait, i3c_master_handle_ibi() is
already a work!
+}
+
+static void tty_i3c_rxwork(struct work_struct *work)
+{
+ struct ttyi3c_port *sport = container_of(work, struct ttyi3c_port, rxwork);
+ struct i3c_priv_xfer xfers;
+ int retry = I3C_TTY_RETRY;
+ u16 status = BIT(0);
+
+ do {
+ memset(&xfers, 0, sizeof(xfers));
+ xfers.data.in = sport->buffer;
+ xfers.len = I3C_TTY_TRANS_SIZE;
+ xfers.rnw = 1;
+
+ if (I3C_TTY_RX_STOP & atomic_read(&sport->status))
Hmm, why not much simpler (and yet atomic) {set,test,clear}_bit()?
+ break;
+
+ i3c_device_do_priv_xfers(sport->i3cdev, &xfers, 1);
+
+ if (xfers.actual_len) {
+ tty_insert_flip_string(&sport->port, sport->buffer, xfers.actual_len);
This can fail.
+ retry = 20;
+ continue;
+ } else {
"} else {" uneeded.
+ status = BIT(0);
+ i3c_device_getstatus_format1(sport->i3cdev, &status);
+ /*
+ * Target side need some time to fill data into fifo. Target side may not
"needs"
+ * have hardware update status in real time. Software update status always
+ * need some delays.
"needs"
+ *
+ * Generally, target side have cicular buffer in memory, it will be moved
"circular" all around.
+ * into FIFO by CPU or DMA. 'status' just show if cicular buffer empty. But
+ * there are gap, espcially CPU have not response irq to fill FIFO in time.
espcially
+ * So xfers.actual will be zero, wait for little time to avoid flood
+ * transfer in i3c bus.
+ */
+ usleep_range(I3C_TTY_YIELD_US, 10 * I3C_TTY_YIELD_US);
+ retry--;
+ }
+
+ } while (retry && (status & BIT(0)));
+
+ tty_flip_buffer_push(&sport->port);
+}
+
+static void tty_i3c_txwork(struct work_struct *work)
+{
+ struct ttyi3c_port *sport = container_of(work, struct ttyi3c_port, txwork);
+ struct circ_buf *circ = &sport->xmit;
+ int cnt = CIRC_CNT(circ->head, circ->tail, UART_XMIT_SIZE);
+ struct i3c_priv_xfer xfers;
+ int retry = I3C_TTY_RETRY;
+ unsigned long flags;
+ int actual;
+ int ret;
+
+ while (cnt > 0 && retry) {
+ xfers.rnw = 0;
+ xfers.len = CIRC_CNT_TO_END(circ->head, circ->tail, UART_XMIT_SIZE);
+ xfers.len = min_t(u16, 16, xfers.len);
+ xfers.data.out = circ->buf + circ->tail;
+
+ ret = i3c_device_do_priv_xfers(sport->i3cdev, &xfers, 1);
+ if (ret) {
+ /*
+ * Target side may not move data out of FIFO. delay can't resolve problem,
+ * just reduce some possiblity. Target can't end I3C SDR mode write
+ * transfer, discard data is reasonable when FIFO overrun.
+ */
+ usleep_range(I3C_TTY_YIELD_US, 10 * I3C_TTY_YIELD_US);
+ retry--;
+ } else {
+ retry = 0;
+ }
+
+ actual = xfers.len;
+
+ circ->tail = (circ->tail + actual) & (UART_XMIT_SIZE - 1);
+
+ if (CIRC_CNT(circ->head, circ->tail, UART_XMIT_SIZE) < WAKEUP_CHARS)
+ tty_port_tty_wakeup(&sport->port);
+
+ cnt = CIRC_CNT(circ->head, circ->tail, UART_XMIT_SIZE);
+ }
+
+ spin_lock_irqsave(&sport->xlock, flags);
+ if (circ->head == circ->tail)
+ complete(&sport->txcomplete);
+ spin_unlock_irqrestore(&sport->xlock, flags);
+}
+
+static int i3c_probe(struct i3c_device *i3cdev)
+{
+ struct ttyi3c_port *port;
+ struct device *tty_dev;
+ struct i3c_ibi_setup req;
+ int minor;
+ int ret;
+
+ port = devm_kzalloc(&i3cdev->dev, sizeof(*port), GFP_KERNEL);
+ if (!port)
+ return -ENOMEM;
+
+ port->i3cdev = i3cdev;
+ port->buffer = devm_kzalloc(&i3cdev->dev, UART_XMIT_SIZE, GFP_KERNEL);
+ if (!port->buffer)
+ return -ENOMEM;
+
+ port->xmit.buf = devm_kzalloc(&i3cdev->dev, UART_XMIT_SIZE, GFP_KERNEL);
+ if (!port->xmit.buf)
+ return -ENOMEM;
You allocate two pages even if you never use the device?
+ dev_set_drvdata(&i3cdev->dev, port);
+
+ req.max_payload_len = 8;
+ req.num_slots = 4;
+ req.handler = &i3c_controller_irq_handler;
+
+ ret = i3c_device_request_ibi(i3cdev, &req);
+ if (ret)
+ return -EINVAL;
+
+ mutex_lock(&i3c_tty_minors_lock);
+ minor = idr_alloc(&i3c_tty_minors, port, 0, I3C_TTY_MINORS, GFP_KERNEL);
+ mutex_unlock(&i3c_tty_minors_lock);
+
+ if (minor < 0)
+ return -EINVAL;
+
+ spin_lock_init(&port->xlock);
+ INIT_WORK(&port->txwork, tty_i3c_txwork);
+ INIT_WORK(&port->rxwork, tty_i3c_rxwork);
+ init_completion(&port->txcomplete);
+
+ port->workqueue = alloc_workqueue("%s", 0, 0, dev_name(&i3cdev->dev));
+ if (!port->workqueue)
+ return -ENOMEM;
+
+ tty_port_init(&port->port);
+ port->port.ops = &i3c_port_ops;
+
+ tty_dev = tty_port_register_device(&port->port, i3c_tty_driver, minor,
+ &i3cdev->dev);
+ if (IS_ERR(tty_dev)) {
+ destroy_workqueue(port->workqueue);
What about tty_port_put() (it drops the idr too)? And free ibi?
+ return PTR_ERR(tty_dev);
+ }
+
+ port->minor = minor;
+
+ return 0;
+}
+
+void i3c_remove(struct i3c_device *dev)
+{
+ struct ttyi3c_port *sport = dev_get_drvdata(&dev->dev);
+
+ tty_port_unregister_device(&sport->port, i3c_tty_driver, sport->minor);
+ cancel_work_sync(&sport->txwork);
+ destroy_workqueue(sport->workqueue);
The same as above.
+}
+
+static struct i3c_driver i3c_driver = {
+ .driver = {
+ .name = "ttyi3c",
+ },
+ .probe = i3c_probe,
+ .remove = i3c_remove,
+ .id_table = i3c_ids,
+};
+
+static int __init i3c_tty_init(void)
+{
+ int ret;
+
+ i3c_tty_driver = tty_alloc_driver(I3C_TTY_MINORS,
+ TTY_DRIVER_REAL_RAW | TTY_DRIVER_DYNAMIC_DEV);
+
+ if (IS_ERR(i3c_tty_driver))
+ return PTR_ERR(i3c_tty_driver);
+
+ i3c_tty_driver->driver_name = "ttyI3C";
+ i3c_tty_driver->name = "ttyI3C";
+ i3c_tty_driver->minor_start = 0,
+ i3c_tty_driver->type = TTY_DRIVER_TYPE_SERIAL,
+ i3c_tty_driver->subtype = SERIAL_TYPE_NORMAL,
+ i3c_tty_driver->init_termios = tty_std_termios;
+ i3c_tty_driver->init_termios.c_cflag = B9600 | CS8 | CREAD | HUPCL |
+ CLOCAL;
+ i3c_tty_driver->init_termios.c_lflag = 0;
+
+ tty_set_operations(i3c_tty_driver, &i3c_tty_ops);
+
+ ret = tty_register_driver(i3c_tty_driver);
+ if (ret) {
+ tty_driver_kref_put(i3c_tty_driver);
+ return ret;
+ }
+
+ ret = i3c_driver_register(&i3c_driver);
+ if (ret) {
+ tty_unregister_driver(i3c_tty_driver);
+ tty_driver_kref_put(i3c_tty_driver);
Use goto + fail paths. And in i3c_probe() too.
+ }
+
+ return ret;
+}
regards,
--
js
suse labs
