Re: lockdep and threaded IRQs (was: ...)

[Peter Zijlstra]

On Mon, 2009-03-02 at 14:09 -0800, David Brownell wrote:
> On Monday 02 March 2009, Peter Zijlstra wrote:
> > > > How so?, its the natural extension of that work.
> > > 
> > > Not the work to shrink the amount of time IRQ latencies
> > > by shrinking the amount of time IRQs are disabled by
> > > IRQ handlers.
> > 
> > Ugh, that's done by pushing work out of the hardirq context,
> 
> That's one of many techniques currently used.
> 
> Tradeoffs don't always favor larger driver updates
> and re-validation though.  Sometimes it's simpler
> to just leverage the reality that "hardirq context"
> does not require using IRQF_DISABLED.

Simpler doesn't mean its a good idea, it opens the door to stack
overruns for one.

Its very unfortunate that people think its a good idea.

> > not by doing silly things like enabling irqs from hardirq context.
> 
> Somehow I'm certain you have NOT analysed every one of the
> thousands of IRQ handlers in various Linux drivers to know
> with certainty that's not the reason IRQ_DISABLED is cleared.
> 
> There are also *other* reasons to leave IRQ_DISABLED clear.

There might be reasons to do so, that doesn't mean its a good idea and
the desired thing to do.

I state that every !IRQF_DISABLED usage is a bug, either due to broken
hardware or broken drivers.

--
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/