Re: [patch 1/3] radix tree: RCU lockless read-side

["Balbir Singh" <bsingharora@gmail.com>]

On 3/11/06, Nick Piggin <nickpiggin@yahoo.com.au> wrote:
> Balbir Singh wrote:
> > <snip>
> >
> >>                if (slot->slots[i]) {
> >>-                       results[nr_found++] = slot->slots[i];
> >>+                       results[nr_found++] = &slot->slots[i];
> >>                        if (nr_found == max_items)
> >>                                goto out;
> >>                }
> >
> >
> > A quick clarification - Shouldn't accesses to slot->slots[i] above be
> > protected using rcu_derefence()?
> >
>
> I think we're safe here -- this is the _address_ of the pointer.
> However, when dereferencing this address in _gang_lookup,
> I think we do need rcu_dereference indeed.
>

Yes, I saw the address operator, but we still derefence "slots" to get
the address.

> Note that _gang_lookup_slot doesn't do this for us, however --
> the caller must do that when dereferencing the pointer to the
> item (eg. see page_cache_get_speculative in 2/3).

Oh! I did not get that far. Will look at the rest of the series

>
> That said, I'm not 100% sure I have the rcu memory barriers in
> the right places (well I'm sure I don't, given the _gang_lookup
> bug you exposed!).

Hmm... Let me look at rcu_torture module and see if I can figure it
out or read the documentation again.

>
> Thanks,
> Nick
>

Warm Regards,
Balbir

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>