On Sun, 12 Jul 2009 21:07:30 -0700 (PDT)
David Miller [off-list ref] wrote:

From: Alan Cox <redacted>
Date: Sun, 12 Jul 2009 23:48:54 +0100

quoted

quoted

The issue was, that the locking code in mkiss was assuming it was only
ever being called in process or bh context.  Fixed by converting the
involved locking code to use irq-safe locks.

Which historically of course was correct until the networking layer
changed behaviour some years ago.

Networking layer?  Surely you mean the TTY layer or USB because I
don't see anything networking in his traces :-)

No I mean the network layer.


Historically it went

		network bh
			transmit packet
				ldisc layer
					tty write method

and lots of code assumed that. Today the network bits get kicked off at
IRQ level.

Most tty stuff now handles it, some of it has been broken with ppp for
ages but its been so long its hardly a regression any more and its dumb
stuff like jtag consoles and i2c serial ports that break.