Serial-Core: adds atomic context debug code.

Andrew,

While porting the USB-Serial layer to the Serial Core API [1], we found that
usb-serial devices must be able to sleep in the get_mctrl() and set_mctrl() callbacks [2]. But, unfortunately, those callbacks are called with a spinlock held.

The solution is to switch from the spinlock to a mutex.

But turns out that we have no sure whether uart_update_mctrl(),
uart_startup() and uart_configure_port() (which calls {get,set}_mctrl()) are called from atomic context or not.

This patch adds might_sleep() calls to them in order to help us to find the
answer.

Please, note that this is just a debug patch to stay in -mm for a while,
ie, it's not a submission for mainline kernel.

[1] http://marc.theaimsgroup.com/?l=linux-usb-devel&m=114921742628790&w=2
[2] http://lkml.org/lkml/2006/6/13/242

diff -purN linux-2.6.17.orig/drivers/serial/serial_core.c linux-2.6.17/drivers/serial/serial_core.c
@@ -128,6 +128,8 @@ uart_update_mctrl(struct uart_port *port
 	unsigned long flags;
 	unsigned int old;
 
+	might_sleep();
+
 	spin_lock_irqsave(&port->lock, flags);
 	old = port->mctrl;
 	port->mctrl = (old & ~clear) | set;

@@ -150,6 +152,8 @@ static int uart_startup(struct uart_stat
 	unsigned long page;
 	int retval = 0;
 
+	might_sleep();
+
 	if (info->flags & UIF_INITIALIZED)
 		return 0;
 

@@ -2055,6 +2059,8 @@ uart_configure_port(struct uart_driver *
 {
 	unsigned int flags;
 
+	might_sleep();
+
 	/*
 	 * If there isn't a port here, don't do anything further.
 	 */

Luiz Fernando N. Capitulino