Skip to contentSkip to conversation

Series short description

PATCH New Alan Cox 1 participant 10 messages Jun 16, 2009

Subsystems (3)
lore
Share & tools
lore

Alan Cox

author
#0 lore raw

cover Series short description

Small fixes to the tty changes to sort out problems people hit, in particular to fix a couple of SMP races Ingo found and properly fix the now rather more visible (but long standing) pty race that James Bottomley encountered.
---

Alan Cox (5):
tty: fix sanity check
pty: Narrow the race on ldisc locking
ldisc: debug aids
ldisc: Make sure the ldisc isn't active when we close it
tty: Fix leaks introduced by the shift to separate ldisc objects

Haavard Skinnemoen (1):
atmel_serial: fix hang in set_termios when crtscts is enabled

Joe Perches (1):
MAINTAINERS: update 8250 section, give Alan Cox a name

Mike Frysinger (1):
tty: fix unused warning when TCGETX is not defined

Sascha Hauer (1):
imx: Check for NULL pointer deref before calling tty_encode_baud_rate

 MAINTAINERS                   |    3 ++
 drivers/char/pty.c            |   53 +++++++++++++++++++++++++++++++----------
 drivers/char/tty_io.c         |    2 ++
 drivers/char/tty_ioctl.c      |    5 ++--
 drivers/char/tty_ldisc.c      |    9 +++++--
 drivers/serial/atmel_serial.c |    8 ++++--
 drivers/serial/imx.c          |   12 +++++----
 7 files changed, 66 insertions(+), 26 deletions(-)
Signature · 3 lines
	"Standards committees don't like hashing.  It looks complicated and
         insufficiently deterministic on an overhead projector."
			-Vern Schryver
author
#1 lore raw

PATCH 1/9 tty: Fix leaks introduced by the shift to separate ldisc objects

From: Alan Cox <[email protected]>

Gold star for the kmemleak detector.

---

 drivers/char/tty_ldisc.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)
drivers/char/tty_ldisc.c +4 -2
index 39c8f86..94b3e06 100644
@@ -148,8 +148,10 @@ static struct tty_ldisc *tty_ldisc_try_get(int disc)
 		}
 	}
 	spin_unlock_irqrestore(&tty_ldisc_lock, flags);
-	if (err)
+	if (err) {
+		kfree(ld);
 		return ERR_PTR(err);
+	}
 	return ld;
 }
 
@@ -262,7 +264,7 @@ const struct file_operations tty_ldiscs_proc_fops = {
  *	@ld: line discipline
  *
  *	Install an instance of a line discipline into a tty structure. The
- *	ldisc must have a reference count above zero to ensure it remains/
+ *	ldisc must have a reference count above zero to ensure it remains.
  *	The tty instance refcount starts at zero.
  *
  *	Locking:
author
#2 lore raw

PATCH 2/9 ldisc: Make sure the ldisc isn't active when we close it

In reply to Alan Cox

From: Alan Cox <[email protected]>

---

 drivers/char/tty_ldisc.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
drivers/char/tty_ldisc.c +2 -0
index 94b3e06..874c248 100644
@@ -793,6 +793,8 @@ void tty_ldisc_hangup(struct tty_struct *tty)
 		/* Avoid racing set_ldisc */
 		mutex_lock(&tty->ldisc_mutex);
 		/* Switch back to N_TTY */
+		tty_ldisc_halt(tty);
+		tty_ldisc_wait_idle(tty);
 		tty_ldisc_reinit(tty);
 		/* At this point we have a closed ldisc and we want to
 		   reopen it. We could defer this to the next open but
author
#3 lore raw

PATCH 3/9 ldisc: debug aids

In reply to Alan Cox

From: Alan Cox <[email protected]>

---

 drivers/char/tty_ldisc.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
drivers/char/tty_ldisc.c +1 -0
index 874c248..a19e935 100644
@@ -207,6 +207,7 @@ static void tty_ldisc_put(struct tty_ldisc *ld)
 	ldo->refcount--;
 	module_put(ldo->owner);
 	spin_unlock_irqrestore(&tty_ldisc_lock, flags);
+	WARN_ON(ld->refcount);
 	kfree(ld);
 }
 
author
#4 lore raw

PATCH 4/9 tty: fix unused warning when TCGETX is not defined

In reply to Alan Cox

From: Mike Frysinger <[email protected]>

If TCGETX is not defined, we end up with this warning:
drivers/char/tty_ioctl.c: In function ‘tty_mode_ioctl’:
drivers/char/tty_ioctl.c:950: warning: unused variable ‘ktermx’

Since the variable is only used in one case statement, push it down to the local case scope.

---

 drivers/char/tty_ioctl.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
drivers/char/tty_ioctl.c +3 -2
index 8116bb1..b24f6c6 100644
@@ -947,7 +947,6 @@ int tty_mode_ioctl(struct tty_struct *tty, struct file *file,
 	void __user *p = (void __user *)arg;
 	int ret = 0;
 	struct ktermios kterm;
-	struct termiox ktermx;
 
 	if (tty->driver->type == TTY_DRIVER_TYPE_PTY &&
 	    tty->driver->subtype == PTY_TYPE_MASTER)
@@ -1049,7 +1048,8 @@ int tty_mode_ioctl(struct tty_struct *tty, struct file *file,
 		return ret;
 #endif
 #ifdef TCGETX
-	case TCGETX:
+	case TCGETX: {
+		struct termiox ktermx;
 		if (real_tty->termiox == NULL)
 			return -EINVAL;
 		mutex_lock(&real_tty->termios_mutex);
@@ -1058,6 +1058,7 @@ int tty_mode_ioctl(struct tty_struct *tty, struct file *file,
 		if (copy_to_user(p, &ktermx, sizeof(struct termiox)))
 			ret = -EFAULT;
 		return ret;
+	}
 	case TCSETX:
 		return set_termiox(real_tty, p, 0);
 	case TCSETXW:
author
#5 lore raw

PATCH 5/9 pty: Narrow the race on ldisc locking

In reply to Alan Cox

From: Alan Cox <[email protected]>

The pty code has always been buggy on its ldisc handling. The recent changes made the window for the race much bigger. Pending fixing it properly which is not at all trivial, at least make the race small again so we don't disrupt other dev work.

---

 drivers/char/pty.c |   53 +++++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 40 insertions(+), 13 deletions(-)
drivers/char/pty.c +40 -13
index 5acd29e..daebe1b 100644
@@ -95,23 +95,34 @@ static void pty_unthrottle(struct tty_struct *tty)
  * a count.
  *
  * FIXME: Our pty_write method is called with our ldisc lock held but
- * not our partners. We can't just take the other one blindly without
- * risking deadlocks.
+ * not our partners. We can't just wait on the other one blindly without
+ * risking deadlocks. At some point when everything has settled down we need
+ * to look into making pty_write at least able to sleep over an ldisc change.
+ *
+ * The return on no ldisc is a bit counter intuitive but the logic works
+ * like this. During an ldisc change the other end will flush its buffers. We
+ * thus return the full length which is identical to the case where we had
+ * proper locking and happened to queue the bytes just before the flush during
+ * the ldisc change.
  */
 static int pty_write(struct tty_struct *tty, const unsigned char *buf,
 								int count)
 {
 	struct tty_struct *to = tty->link;
-	int	c;
+	struct tty_ldisc *ld;
+	int c = count;
 
 	if (!to || tty->stopped)
 		return 0;
-
-	c = to->receive_room;
-	if (c > count)
-		c = count;
-	to->ldisc->ops->receive_buf(to, buf, NULL, c);
-
+	ld = tty_ldisc_ref(to);
+
+	if (ld) {
+		c = to->receive_room;
+		if (c > count)
+			c = count;
+		ld->ops->receive_buf(to, buf, NULL, c);
+		tty_ldisc_deref(ld);
+	}
 	return c;
 }
 
@@ -145,14 +156,23 @@ static int pty_write_room(struct tty_struct *tty)
 static int pty_chars_in_buffer(struct tty_struct *tty)
 {
 	struct tty_struct *to = tty->link;
-	int count;
+	struct tty_ldisc *ld;
+	int count = 0;
 
 	/* We should get the line discipline lock for "tty->link" */
-	if (!to || !to->ldisc->ops->chars_in_buffer)
+	if (!to)
+		return 0;
+	/* We cannot take a sleeping reference here without deadlocking with
+	   an ldisc change - but it doesn't really matter */
+	ld = tty_ldisc_ref(to);
+	if (ld == NULL)
 		return 0;
 
 	/* The ldisc must report 0 if no characters available to be read */
-	count = to->ldisc->ops->chars_in_buffer(to);
+	if (ld->ops->chars_in_buffer)
+		count = ld->ops->chars_in_buffer(to);
+
+	tty_ldisc_deref(ld);
 
 	if (tty->driver->subtype == PTY_TYPE_SLAVE)
 		return count;
@@ -182,12 +202,19 @@ static void pty_flush_buffer(struct tty_struct *tty)
 {
 	struct tty_struct *to = tty->link;
 	unsigned long flags;
+	struct tty_ldisc *ld;
 
 	if (!to)
 		return;
+	ld = tty_ldisc_ref(to);
+
+	/* The other end is changing discipline */
+	if (!ld)
+		return;
 
-	if (to->ldisc->ops->flush_buffer)
+	if (ld->ops->flush_buffer)
 		to->ldisc->ops->flush_buffer(to);
+	tty_ldisc_deref(ld);
 
 	if (to->packet) {
 		spin_lock_irqsave(&tty->ctrl_lock, flags);
author
#6 lore raw

PATCH 6/9 tty: fix sanity check

In reply to Alan Cox

From: Alan Cox <[email protected]>

The WARN_ON() that was added to tty_reopen can be triggered in the specific case of a hangup occurring during a re-open of a tty which is not in the middle of being otherwise closed.

In that case however the WARN() is bogus as we don't hold the neccessary locks to make a correct decision.

The case we should be checking is "if the ldisc is not changing and reopen is occuring". We could drop the WARN_ON but for the moment the debug is more valuable even if it means taking a mutex as it will find any other cases.

---

 drivers/char/tty_io.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
drivers/char/tty_io.c +2 -0
index 939e198..a3afa0c 100644
@@ -1263,7 +1263,9 @@ static int tty_reopen(struct tty_struct *tty)
 	tty->count++;
 	tty->driver = driver; /* N.B. why do this every time?? */
 
+	mutex_lock(&tty->ldisc_mutex);
 	WARN_ON(!test_bit(TTY_LDISC, &tty->flags));
+	mutex_unlock(&tty->ldisc_mutex);
 
 	return 0;
 }
author
#7 lore raw

PATCH 7/9 MAINTAINERS: update 8250 section, give Alan Cox a name

In reply to Alan Cox

From: Joe Perches <[email protected]>

---

 MAINTAINERS |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
MAINTAINERS +2 -1
index 2cb7566..09f6b3e 100644
@@ -157,9 +157,10 @@ S:	Maintained
 F:	drivers/net/r8169.c
 
 8250/16?50 (AND CLONE UARTS) SERIAL DRIVER
+P:	Alan Cox
+M:	[email protected]
 L:	[email protected]
 W:	http://serial.sourceforge.net
-M:	[email protected]
 S:	Odd Fixes
 F:	drivers/serial/8250*
 F:	include/linux/serial_8250.h
author
#8 lore raw

PATCH 8/9 atmel_serial: fix hang in set_termios when crtscts is enabled

In reply to Alan Cox

From: Haavard Skinnemoen <[email protected]>

After enabling hardware flow control, any subsequent termios call may hang waiting for the transmitter to drain. This appears to be caused by a busy-loop in set_termios() waiting for the transmitter to become empty, which may take a very long time (or hang indefinitely) if the device at the other end is blocking us.

A quick look through the tty and serial_core code indicates that any necessary flushing (which is optional) has already been done at this point, so there's no need for the driver to flush the transmitter on its own.

Fix it by removing the busy-loop altogether.

---

 drivers/serial/atmel_serial.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)
drivers/serial/atmel_serial.c +5 -3
index b3497d7..338b15c 100644
@@ -1104,11 +1104,13 @@ static void atmel_set_termios(struct uart_port *port, struct ktermios *termios,
 	/* update the per-port timeout */
 	uart_update_timeout(port, termios->c_cflag, baud);
 
-	/* save/disable interrupts and drain transmitter */
+	/*
+	 * save/disable interrupts. The tty layer will ensure that the
+	 * transmitter is empty if requested by the caller, so there's
+	 * no need to wait for it here.
+	 */
 	imr = UART_GET_IMR(port);
 	UART_PUT_IDR(port, -1);
-	while (!(UART_GET_CSR(port) & ATMEL_US_TXEMPTY))
-		cpu_relax();
 
 	/* disable receiver and transmitter */
 	UART_PUT_CR(port, ATMEL_US_TXDIS | ATMEL_US_RXDIS);
author
#9 lore raw

PATCH 9/9 imx: Check for NULL pointer deref before calling tty_encode_baud_rate

In reply to Alan Cox

From: Sascha Hauer <[email protected]>

---

 drivers/serial/imx.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)
drivers/serial/imx.c +7 -5
index 285b414..5d7b58f 100644
@@ -924,11 +924,13 @@ imx_set_termios(struct uart_port *port, struct ktermios *termios,
 	rational_best_approximation(16 * div * baud, sport->port.uartclk,
 		1 << 16, 1 << 16, &num, &denom);
 
-	tdiv64 = sport->port.uartclk;
-	tdiv64 *= num;
-	do_div(tdiv64, denom * 16 * div);
-	tty_encode_baud_rate(sport->port.info->port.tty,
-		(speed_t)tdiv64, (speed_t)tdiv64);
+	if (port->info && port->info->port.tty) {
+		tdiv64 = sport->port.uartclk;
+		tdiv64 *= num;
+		do_div(tdiv64, denom * 16 * div);
+		tty_encode_baud_rate(sport->port.info->port.tty,
+				(speed_t)tdiv64, (speed_t)tdiv64);
+	}
 
 	num -= 1;
 	denom -= 1;