[lkml]   [2014]   [Nov]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2 9/9] netfilter: Replace smp_read_barrier_depends() with lockless_dereference()
On Fri, Nov 21, 2014 at 7:05 PM, Eric Dumazet <> wrote:
> On Fri, 2014-11-21 at 16:57 -0500, Pranith Kumar wrote:
>> Hi Eric,
>> Thanks for looking at this patch.
>> I've been scratching my head since morning trying to find out what was
>> so obviously wrong with this patch. Alas, I don't see what you do.
>> Could you point it out and show me how incompetent I am, please?
>> Thanks!
> Well, even it the code is _not_ broken, I don't see any value with this
> patch.

Phew. Not being broken itself is a win :)

> If I use git blame on current code, line containing
> smp_read_barrier_depends() exactly points to the relevant commit [1]

And that is an opinion I will respect. I don't want to muck the git
history where it is significant.

This effort is to eventually replace the uses of
smp_read_barrier_depends() and to use either rcu or
lockless_dereference() as documented in memory-barriers.txt.

> After your change, it will point to some cleanup, which makes little
> sense to me, considering you did not change the smp_wmb() in
> xt_replace_table().

That does not need to change as it is fine as it is. It still pairs
with the smp_read_barrier_depends() in lockless_dereference().

> I, as a netfilter contributor would like to keep current code as is,
> because it is how I feel safe with it.
> We have a proliferation of interfaces, but this does not help to
> understand the issues and code maintenance.
> smp_read_barrier_depends() better documents the read barrier than
> lockless_dereference().

I think this is a matter of opinion. But in the current effort I've
seen cases where it is not clear what the barrier is actually
guaranteeing. I am glad that the current code is not one of those and
it has reasonable comments.

lockless_dereference() on the other hand makes the dependency explicit.

> The point of having a lock or not is irrelevant here.
> [1]


 \ /
  Last update: 2014-11-22 03:01    [W:0.075 / U:36.752 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site