--
On Mon, 30 Jul 2007, Ingo Molnar wrote:

quoted

From: Steven Rostedt <rostedt@goodmis.org>

I think this was sent before, and it did cause problems before. Would
there be *any* reason to have non-threaded softirqs but threaded
hardirqs. I can see lots of issues with that.

please elaborate in precise terms: what issues can you see?

Hi Ingo,

I don't remember the exact details, I can try to find the thread. But I
remember someone was having their system lock up strangly. We later found
that they had softirqs as normal softirqs and interrupts as threads.  I
think there was some driver that didn't expect the softirq to preempt the
irq handler.  Perhaps the softirq was using spin_lock_irq while the irq
thread was just using spin_lock, which I can see as being something
normal.

The standard Linux does not expect an interrupt to be preempted by a
softirq, and with interrupts as threads but not softirqs, I can see that
happening a lot.

-- Steve