On Tue, 22 Apr 2008, Jens Axboe wrote:
> >
> > You forgot to free the "data" here? The waiter must also free the object,
> > since now the callee does not.
>
> The ipi interrupt handler does that, see kfree() in
> generic_smp_call_function_single_interrupt() or call_func_data_free() in
> generic_smp_call_function_interrupt().
Hell no, it does *not*.
Doing that for the waiting case would be a *huge* bug, since the waiter
needs to wait until the flag is clear - and if the waitee free's the
allocation, that will never happen.
So the rule *must* be:
- waiter frees
- ipi interrupt frees non-waiting ones.
because anything else cannot work.
And you must have known that, because the code you pointed me to does
*not* free the data at all. It just clears the FLAG_WAIT flag:
+ if (data->csd.flags & CSD_FLAG_WAIT) {
+ smp_wmb();
+ data->csd.flags &= ~CSD_FLAG_WAIT;
+ } else
+ call_func_data_free(data);
So please think about this some more.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
