work_on_cpu(): rewrite it to create a kernel thread on demand
authorAndrew Morton <akpm@linux-foundation.org>
Thu, 9 Apr 2009 15:50:37 +0000 (09:50 -0600)
committerRusty Russell <rusty@rustcorp.com.au>
Thu, 9 Apr 2009 00:20:37 +0000 (09:20 +0930)
commit6b44003e5ca66a3fffeb5bc90f40ada2c4340896
treed9dce0a39b5e66d8e760344a51ffb6de9594cd2f
parent1c99315bb36b5d776210546d438ca928dc9b1f22
work_on_cpu(): rewrite it to create a kernel thread on demand

Impact: circular locking bugfix

The various implemetnations and proposed implemetnations of work_on_cpu()
are vulnerable to various deadlocks because they all used queues of some
form.

Unrelated pieces of kernel code thus gained dependencies wherein if one
work_on_cpu() caller holds a lock which some other work_on_cpu() callback
also takes, the kernel could rarely deadlock.

Fix this by creating a short-lived kernel thread for each work_on_cpu()
invokation.

This is not terribly fast, but the only current caller of work_on_cpu() is
pci_call_probe().

It would be nice to find some other way of doing the node-local
allocations in the PCI probe code so that we can zap work_on_cpu()
altogether.  The code there is rather nasty.  I can't think of anything
simple at this time...

Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
kernel/workqueue.c