sllin: update for kernels > 3.11

I'm experimenting with sllin ( https://rtime.felk.cvut.cz/can/lin-bus/ ), and I've found that it doesn't compile on linux kernels newer than 3.11, so I've patched it to compile and (hopefully) run on newer kernels.

I've verified that these changes compile and the resulting module loads on the 4.2.0 kernel on my Ubuntu machine, but I don't have LIN hardware yet and haven't tried attaching the LIN serial line discipline to a serial port.

In the event that this driver is no longer maintained, I'm also hosting my copy of the source on github: https://github.com/trainman419/linux-lin

Author: Austin Hendrix <[email protected]>
Date:   Mon Apr 25 10:41:13 2016 -0700

Update for changes in recent linux kernels

index 8d7ad69..2db896f 100644

@@ -210,7 +210,11 @@ static int sltty_change_speed(struct tty_struct *tty, unsigned speed)
     struct ktermios old_termios, termios;
     int cflag;
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 12, 0)
     mutex_lock(&tty->termios_mutex);
+#else
+    down_write(&tty->termios_rwsem);
+#endif
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 7, 0)
     old_termios = termios = *(tty->termios);

@@ -238,7 +242,11 @@ static int sltty_change_speed(struct tty_struct *tty, unsigned speed)
     if (tty->ops->set_termios)
         tty->ops->set_termios(tty, &old_termios);
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 12, 0)
     mutex_unlock(&tty->termios_mutex);
+#else
+    up_write(&tty->termios_rwsem);
+#endif
 
     return 0;
 }

@@ -311,7 +319,12 @@ static void sllin_write_wakeup(struct tty_struct *tty)
             return;    /* ongoing concurrent processing */
 
         clear_bit(SLF_TXBUFF_RQ, &sl->flags);
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
         smp_mb__after_clear_bit();
+#else
+        smp_mb__after_atomic();
+#endif
 
         if (sl->lin_state != SLSTATE_BREAK_SENT)
             remains = sl->tx_lim - sl->tx_cnt;

@@ -325,7 +338,11 @@ static void sllin_write_wakeup(struct tty_struct *tty)
             remains -= actual;
         }
         clear_bit(SLF_TXBUFF_INPR, &sl->flags);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
         smp_mb__after_clear_bit();
+#else
+        smp_mb__after_atomic();
+#endif
 
     } while (unlikely(test_bit(SLF_TXBUFF_RQ, &sl->flags)));
 

@@ -838,7 +855,11 @@ static int sllin_send_tx_buff(struct sllin *sl)
             return 0;    /* ongoing concurrent processing */
 
         clear_bit(SLF_TXBUFF_RQ, &sl->flags);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
         smp_mb__after_clear_bit();
+#else
+        smp_mb__after_atomic();
+#endif
 
 #ifdef BREAK_BY_BAUD
         if (sl->lin_state != SLSTATE_BREAK_SENT)

@@ -872,7 +893,11 @@ static int sllin_send_tx_buff(struct sllin *sl)
                 sl->tx_cnt, remains);
 
         clear_bit(SLF_TXBUFF_INPR, &sl->flags);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 18, 0)
         smp_mb__after_clear_bit();
+#else
+        smp_mb__after_atomic();
+#endif
 
     } while (unlikely(test_bit(SLF_TXBUFF_RQ, &sl->flags)));
 

@@ -1337,7 +1362,12 @@ static struct sllin *sll_alloc(dev_t line)
         char name[IFNAMSIZ];
         sprintf(name, "sllin%d", i);
 
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(3, 17, 0))
         dev = alloc_netdev(sizeof(*sl), name, sll_setup);
+#else
+        dev = alloc_netdev(sizeof(*sl), name, NET_NAME_UNKNOWN, sll_setup);
+#endif
+
         if (!dev)
             return NULL;
         dev->base_addr  = i;

Hello Austin,

I have pushed your patch to our (CTU) repository. The changes looks good and your report that code build should be enough evidence that these changes are going in the right direction for newer kernels.

We need to find more time to recheck driver functionality but we do not have any LIN project running at this moment and are too busy with other projects. But next round of testing is no my long long TODO list.

The simple RS-232 to LIN converter hardware is described in the RTLWS paper

https://rtime.felk.cvut.cz/can/sllin-rtlws14-paper.pdf

There are some other related document on our CAN and LIN related pages

https://rtime.felk.cvut.cz/can/lin-bus/

The proper converter from TTL or RS-232 UART to LIN could be build with use of some standard compliant LIN transceiver chip.

Generally, code should work with all kinds of UART drivers except for the case when slLIN operates as device and is asked to send response. In such case there can be problem that UART with FIFO usually waits for four character times before it reports received characters if FIFO does not reach Rx FIFO trigger level. If the other side LIN master is strict and response does not start in two character times then it reports error. The trigger level can be set to single character for many embedded UARTs or Rx FIFO can be disabled completely. But there there has been problem in the past that Linux termios based serial ports did not expose interface to setup trigger level only limited trigger level to one for baudrates < 2400.

But when I analyze actual standard PC/8250 based driver code

drivers/tty/serial/8250/8250_port.c

        if (up->capabilities & UART_CAP_FIFO && port->fifosize > 1) {
                /* NOTE: If fifo_bug is not set, a user can set RX_trigger. */
                if ((baud < 2400 && !up->dma) || up->fifo_bug) {
                        up->fcr &= ~UART_FCR_TRIGGER_MASK;
                        up->fcr |= UART_FCR_TRIGGER_1;
                }
        }

static int do_set_rxtrig(struct tty_port *port, unsigned char bytes)

static DEVICE_ATTR(rx_trig_bytes, S_IRUSR | S_IWUSR | S_IRGRP,

                   serial8250_get_attr_rx_trig_bytes,
                   serial8250_set_attr_rx_trig_bytes);

There is defined new attribute to select trigger level from userspace, so this problem should be easily eliminated by setting trigger level to one character

echo 1 > /sys/class/tty/ttyS0/rx_trig_bytes

We have used hack in the past to modify port trigger level, now it is much easier.

There are alternatives for some other hardware. Look for device attributes or module parameters with rx_trigr in the name or rx_timeout

  MODULE_PARM_DESC(rx_trigger_level, "Rx trigger level, 1-63 bytes");

  MODULE_PARM_DESC(rx_timeout, "Rx timeout, 1-255");

It would be great if you can report results of your testing.

Best wishes,

Pavel