On Thu, 7 Dec 2006, Andrew Morton wrote:
>
> We _can_ trust it in the context of
>
> void flush_work(struct work_struct *work)
Yes, but the way the bits are defined, the "pending" bit is not meaningful
as a synchronization event, for example - because _other_ users can't
trust it once they've dispatched the function. So even in the synchronous
run/flush_scheduled_work() kind of situation, you end up having to work
with the fact that nobody _else_ can rely on the data structures, and that
they are designed to work that way..
> ho-hum. I'll take a look at turning that into something which compiles,
> then I'll convert a few oft-used flush_scheduled_work() callers over to use
> it. To do this on a sensible timescale perhaps means that we should export
> current_is_keventd(), get the howling hordes off our backs.
Well, I simply committed my work that doesn't guarantee synchronization -
the synchronization can now be added in kernel/workqueue.c any way we
want. It's better than what we used to have, for sure, in both compiling
and solving the practical problem, but also as a "go forward" point.
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
