Skip to contentSkip to conversation

8250: Fix oops from setserial

PATCH In review Alan Cox 2 participants 6 messages May 28, 2009 – May 30, 2009

lore
Share & tools
lore

Alan Cox

author
#0 lore raw

PATCH 1/1 8250: Fix oops from setserial

From: Alan Cox <[email protected]>

[Impact: ok I just put this here to wind up Linus]

If you setserial a port which has never been initialised we change the type but don't update the I/O method pointers. The same problem is true if you change the io type of a port - but nobody ever does that so nobody noticed!

Remember the old type and when attaching if the type has changed reload the port accessor pointers. We can't do it blindly as some 8250 drivers load custom accessors and we must not stomp those.

Closes-bug: #13367

---

 drivers/serial/8250.c |   15 +++++++++++++++
 1 files changed, 15 insertions(+), 0 deletions(-)
drivers/serial/8250.c +15 -0
index b4b3981..a0127e9 100644
@@ -137,6 +137,7 @@ struct uart_8250_port {
 	unsigned char		mcr;
 	unsigned char		mcr_mask;	/* mask of user bits */
 	unsigned char		mcr_force;	/* mask of forced bits */
+	unsigned char		cur_iotype;	/* Running I/O type */
 
 	/*
 	 * Some bits in registers are cleared on a read, so they must
@@ -471,6 +472,7 @@ static void io_serial_out(struct uart_port *p, int offset, int value)
 
 static void set_io_from_upio(struct uart_port *p)
 {
+	struct uart_8250_port *up = (struct uart_8250_port *)p;
 	switch (p->iotype) {
 	case UPIO_HUB6:
 		p->serial_in = hub6_serial_in;
@@ -509,6 +511,8 @@ static void set_io_from_upio(struct uart_port *p)
 		p->serial_out = io_serial_out;
 		break;
 	}
+	/* Remember loaded iotype */
+	up->cur_iotype = p->iotype;
 }
 
 static void
@@ -1937,6 +1941,9 @@ static int serial8250_startup(struct uart_port *port)
 	up->capabilities = uart_config[up->port.type].flags;
 	up->mcr = 0;
 
+	if (up->port.iotype != up->cur_iotype)
+		set_io_from_upio(port);
+
 	if (up->port.type == PORT_16C950) {
 		/* Wake up and initialize UART */
 		up->acr = 0;
@@ -2563,6 +2570,9 @@ static void serial8250_config_port(struct uart_port *port, int flags)
 	if (ret < 0)
 		probeflags &= ~PROBE_RSA;
 
+	if (up->port.iotype != up->cur_iotype)
+		set_io_from_upio(port);
+
 	if (flags & UART_CONFIG_TYPE)
 		autoconfig(up, probeflags);
 	if (up->port.type != PORT_UNKNOWN && flags & UART_CONFIG_IRQ)
@@ -2671,6 +2681,11 @@ serial8250_register_ports(struct uart_driver *drv, struct device *dev)
 {
 	int i;
 
+	for (i = 0; i < nr_uarts; i++) {
+		struct uart_8250_port *up = &serial8250_ports[i];
+		up->cur_iotype = 0xFF;
+	}
+
 	serial8250_isa_init_ports();
 
 	for (i = 0; i < nr_uarts; i++) {
#1 lore raw
Quoted 8 lines · On Thu, 28 May 2009 14:01:35 +0100 Alan Cox <[email protected]> wrote:

From: Alan Cox <[email protected]>

[Impact: ok I just put this here to wind up Linus]

If you setserial a port which has never been initialised we change the type but don't update the I/O method pointers. The same problem is true if you change the io type of a port - but nobody ever does that so nobody noticed!

Quoted 5 lines

Remember the old type and when attaching if the type has changed reload the port accessor pointers. We can't do it blindly as some 8250 drivers load custom accessors and we must not stomp those.

Closes-bug: #13367

Please quote the full bug URL.

a) for consistency

b) because there are multiple bug tracking systems out there:

y:/usr/src/git26> git-log|grep bugzilla.redhat|wc -l 123

drivers/serial/8250.c static void io_serial_out(struct uart_port *p, int offset, int value) · line 475

Quoted 3 lines

static void set_io_from_upio(struct uart_port *p) {

drivers/serial/8250.c +1 -0
+	struct uart_8250_port *up = (struct uart_8250_port *)p;

container_of() is nicer, IMO. It's clearer, and doesn't require that the uart_port be the fist member.

author
#2 lore raw
Quoted 7 lines

drivers/serial/8250.c static void io_serial_out(struct uart_port *p, int offset, int value) · line 475

Quoted 4 lines

static void set_io_from_upio(struct uart_port *p) {

drivers/serial/8250.c +1 -0
+	struct uart_8250_port *up = (struct uart_8250_port *)p;

container_of() is nicer, IMO. It's clearer, and doesn't require that the uart_port be the fist member.

See the rest of the driver - it was written that way years ago and this is just following the existing design.

#3 lore raw
Quoted 10 lines · On Sat, 30 May 2009 09:25:57 +0100 Alan Cox <[email protected]> wrote:
Quoted 7 lines

drivers/serial/8250.c static void io_serial_out(struct uart_port *p, int offset, int value) · line 475

Quoted 4 lines

static void set_io_from_upio(struct uart_port *p) {

drivers/serial/8250.c +1 -0
+	struct uart_8250_port *up = (struct uart_8250_port *)p;

container_of() is nicer, IMO. It's clearer, and doesn't require that the uart_port be the fist member.

See the rest of the driver - it was written that way years ago and this is just following the existing design.

Sure, huge numbers of drivers use the cast.

But I've never seen past sinnings as being a reason to continue sinning. Doing new code the right way doesn't result in worse code, and reduces the chances of someone copying and pasting the wrong way.

author
#4 lore raw
Quoted 5 lines

Sure, huge numbers of drivers use the cast.

But I've never seen past sinnings as being a reason to continue sinning. Doing new code the right way doesn't result in worse code, and reduces the chances of someone copying and pasting the wrong way.

As I said please read the rest of the driver. It would ridiculous to have one dereference done differently to the rest in several thousand lines of code.

It's not a new driver, it's simply pasting a line from one place to another to update a field.

#5 lore raw
Quoted 7 lines · On Sat, 30 May 2009 11:46:32 +0100 Alan Cox <[email protected]> wrote:
Quoted 5 lines

Sure, huge numbers of drivers use the cast.

But I've never seen past sinnings as being a reason to continue sinning. Doing new code the right way doesn't result in worse code, and reduces the chances of someone copying and pasting the wrong way.

As I said please read the rest of the driver.

I'm not the reading-averse one here :(

Quoted 3 lines

It would ridiculous to have
one dereference done differently to the rest in several thousand lines of code.

Actually, it would be typical. Many many drivers which originally did something consistently wrong have, over time, grown to contain a mix of right and wrong. It's one of the costs of doing things wrong.

Quoted 2 lines

It's not a new driver, it's simply pasting a line from one place to another to update a field.

You already said all this.

Look, it's a single silly line, but there's a non-trivial point here. It comes up quite regularly and I always encourage people to do things the right way rather than matching the existing wrong code. Because, when you think about it, there's really no merit in having consistently wrong code. A mix of right and wrong is better than 100% wrong.

If it results in inconsistent-looking code then that's good. It may result in some drive-by developer noticing the inconsistency and fixing everything up.

Have a think about it.