Re: PROBLEM: kernel oops with g_serial USB gadget on 2.6.30

[David Brownell <david-b@pacbell.net>]

On Monday 22 June 2009, Marek Szyprowski wrote:
> 
> > This is just a guess...  But there's a good possibility that the oops
> > was caused by recent changes to the serial layer which have not been
> > propagated through to the g_serial driver.
> 
> How recent these changes are? I did a test on another ARM-based Linux 
> platform with old 2.6.28 kernel and the result was exactly the same as 
> above...

Just for the record, the reworked g_serial code merged in 2.6.27
and was mostly developed on 2.6.25 and 2.6.26 ... and it included
a lot of stress testing.  No such mutex_lock() in_irq() bug showed
up at that time.  And that was running with all practical kernel
debug options, so it should have showed up if it were that easy.

I do however recall turning up several regressions in how "sparse"
lock checking behaved.  As in, it broke when faced with common
idioms like needing to temporarily drop a lock deep in a call stack.

Now, the serial layer has been getting a *LONG* overdue incremental
overhaul since before that started.  So there's been plenty of time
for incompatible changes to sneak in; I believe Alan Cox focuses on
host side things, out of defensive necessity.

Like, oh, changing a spinlock to a mutex.   You might change the
low_latency setting and review how that's now supposed to behave.

- Dave