On Sat, Feb 24, 2007 at 10:37:44PM -0800, David Miller wrote:

From: Ingo Molnar <redacted>
Date: Sun, 25 Feb 2007 07:27:47 +0100

quoted

* Paul E. McKenney [off-list ref] wrote:

quoted

I got the following running stock 2.6.20-rt8 on an 4-CPU 1.8GHz 
Opteron box.  The machine continued to run a few rounds of kernbench 
and LTP. Looks a bit scary -- a tasklet was "stolen" from 
__tasklet_action().

Thoughts?  In the meantime, kicking it off again to see if it repeats.

quoted

BUG: at kernel/softirq.c:559 __tasklet_action()

this seems to happen very sporadically. Seems to happen more likely on 
hyperthreading CPUs. It is very likely caused by the 
redesign-tasklet-locking-to-be-sane patch below - which is a quick hack 
of mine from early -rt days. Can you see any obvious bug in it? The 
cmpxchg logic is certainly a bit ... tricky, locking-wise.

Ingo, please don't use cmpxchg() in generic code, we support several
processors that simply cannot do it.

OK, I will bite...

Why doesn't the traditional hash table of locks work here?  Use the
cache-line address as input to the hash function, take the corresponding
lock, do the compare-and-exchange by hand, and then release the lock.
What am I missing here?  Address aliasing do to memory being mapped into
multiple locations or something?  (In that case, use only the portion
of the address within the page, right?)

I will agree that cmpxchg() has been abused pretty thoroughly in
some venues, but it does have legitimate uses.

						Thanx, Paul

Instead of saying "it's just something special in -rt for now", take
it out now so that what you do eventually push upstream does get
tested.