Re: [2.6.9-rc3-bk1] Sleeping function called from invalid context

[Linus Torvalds]

On Fri, 1 Oct 2004, Andrew Morton wrote:
> 
> In rc3 change_termios() was altered so that it calls
> (*tty->driver->set_termios)() under spin_lock_irqsave(tty_termios_lock).
> 
> But ftdi_set_termios() wants to perform USB I/O, which involves sleeping
> allocations all over the place.

Alan, my first reaction was "that's just a buggy driver", but looking at 
the reason for it (telling the device at the other end to change it's 
speed), it's clearly not the driver. The new spinlock is just wrong.

We might be able to change it to a per-tty semaphore instead, which sounds 
reasonably safe, but in the meantime I don't think we have any other 
choice than to just leave the driver/ldisc routines unprotected. They 
should be just reading the state, after all.

I'm committing this temporary fix..

		Linus
----
===== drivers/char/tty_ioctl.c 1.13 vs edited =====
--- 1.13/drivers/char/tty_ioctl.c	2004-09-29 14:27:20 -07:00
+++ edited/drivers/char/tty_ioctl.c	2004-10-02 12:06:40 -07:00
@@ -145,6 +145,13 @@
 			wake_up_interruptible(&tty->link->read_wait);
 		}
 	}
+
+	/*
+	 * Fixme! We should really try to protect the driver and ldisc
+	 * termios usage too. But they need to be able to sleep, so
+	 * the global termios spinlock is not the right thing.
+	 */
+	spin_unlock_irqrestore(&tty_termios_lock, flags);
 	   
 	if (tty->driver->set_termios)
 		(*tty->driver->set_termios)(tty, &old_termios);
@@ -155,7 +162,6 @@
 			(ld->set_termios)(tty, &old_termios);
 		tty_ldisc_deref(ld);
 	}
-	spin_unlock_irqrestore(&tty_termios_lock, flags);
 }
 
 static int set_termios(struct tty_struct * tty, void __user *arg, int opt)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/