Re: [PATCH] Export current_is_keventd() for libphy

[David Howells <dhowells@redhat.com>]

Linus Torvalds <torvalds@osdl.org> wrote:

> How about something like this?

At first glance, this looks reasonable.

It also looks like it should be used to replace a lot of the
cancel_delayed_work() calls that attempt to cancel _undelayed_ work items.
That would allow a number of work items to be downgraded from delayed_work to
work_struct.

Also, the name "run_scheduled_work" sort of suggests that the work *will* be
run regardless of whether it was pending or not.  Given the confusion over
cancel_delayed_work(), I imagine this will rain confusion too.

+	if (get_wq_data(work) == cwq
+	    && test_bit(WORK_STRUCT_PENDING, &work->management)

I wonder if those can be combined, perhaps:

+	if ((work->management & ~WORK_STRUCT_NOAUTOREL) ==
+	    ((unsigned long) cwq | (1 << WORK_STRUCT_PENDING))

Otherwise for i386 the compiler can't combine them because test_bit() is done
with inline asm.

And:

+		if (!test_bit(WORK_STRUCT_PENDING, &work->management))

Should possibly be:

+		if (!work_pending(work))

David