Re: About the fixes of /drivers/serial/8250.C in 2.6.17-rc6 for avoiding habbg-up

Re: About the fixes of /drivers/serial/8250.C in 2.6.17-rc6 for avoiding habbg-up

[Alan Cox]

Ar Llu, 2006-06-19 am 12:03 +0900, ysgrifennodd gouji:
> In /drivers/serial/8250.C of 2.6.17-rc6,
> 
> these fixes are adapted for avoinding  the problem of hang-up
> while TTY write and console write from kernel conflicted.

Yes, there is a bug in this version that was not in the one I submitted,
someone added an improvement.

> +   if (oops_in_progress) {
> +      locked = spin_trylock_irqsave(&up->port.lock, flags);
> +   } else
> +      spin_lock_irqsave(&up->port.lock, flags);
> +

It could always use spin_trylock_irqsave(). The oops in progress
optimisation makes some sense initially but there are many console
printk users that can occur during serial I/O in exceptional cases.

It's not an easy problem to solve with the current serial locking.

Alan

Re: About the fixes of /drivers/serial/8250.C in 2.6.17-rc6 for avoiding habbg-up

[Russell King]

On Mon, Jun 19, 2006 at 10:23:14AM +0100, Alan Cox wrote:
> Ar Llu, 2006-06-19 am 12:03 +0900, ysgrifennodd gouji:
> > In /drivers/serial/8250.C of 2.6.17-rc6,
> > 
> > these fixes are adapted for avoinding  the problem of hang-up
> > while TTY write and console write from kernel conflicted.
> 
> Yes, there is a bug in this version that was not in the one I submitted,
> someone added an improvement.

I disagree - in the non-oops_in_progress case, your version and the
merged version are 100% identical - see

	http://marc.theaimsgroup.com/?l=linux-kernel&m=114657841432447&w=2

However, you never responded to my answers to your two questions in
that email, which came with the patch which was merged.

> > +   if (oops_in_progress) {
> > +      locked = spin_trylock_irqsave(&up->port.lock, flags);
> > +   } else
> > +      spin_lock_irqsave(&up->port.lock, flags);
> > +
> 
> It could always use spin_trylock_irqsave().  The oops in progress
> optimisation makes some sense initially but there are many console
> printk users that can occur during serial I/O in exceptional cases.
> 
> It's not an easy problem to solve with the current serial locking.

I don't have the initial email from ysgrifennodd gouji, and neither
do the lkml archives.  What's the problem?

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: About the fixes of /drivers/serial/8250.C in 2.6.17-rc6 for avoiding habbg-up

[Alan Cox]

Ar Llu, 2006-06-19 am 10:32 +0100, ysgrifennodd Russell King:
> > Yes, there is a bug in this version that was not in the one I submitted,
> > someone added an improvement.
> 
> I disagree - in the non-oops_in_progress case, your version and the
> merged version are 100% identical - see

So they are, sorry the one that differs was the original Red Hat
Enterprise Linux patch.

> I don't have the initial email from ysgrifennodd gouji, and neither
> do the lkml archives.  What's the problem?

Sysrq for one