Re: [PATCH net-next 0/6] net: dsa: push switchdev prepare phase in FDB ops

[Vivien Didelot <vivien.didelot@savoirfairelinux.com>]

Hi Jiri, David,

On Oct. Thursday 08 (41) 03:47 PM, Jiri Pirko wrote:
> Thu, Oct 08, 2015 at 03:32:51PM CEST, vivien.didelot@savoirfairelinux.com wrote:
> >Hi David,
> >
> >On Oct. Thursday 08 (41) 05:28 AM, David Miller wrote:
> >> From: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
> >> Date: Wed,  7 Oct 2015 19:48:25 -0400
> >> 
> >> > The first 3 patches removes the dsa.h include from linux/netdevice.h, which
> >> > broke the inclusion of switchdev.h in dsa.h.
> >> 
> >> I still don't agree with bloating up struct netdevice just to deal with
> >> an include file ordering issue, sorry.
> >
> >Yes, I just saw your reply on the first version. I will resend the
> >patchset with the forward declarations instead.
> >
> >But looking at the issue that Jiri and the kbuild bot pointed out
> >earlier in the thread, we must agree that having the DSA header in
> >netdevice.h is wrong.
> >
> >There are 2 points to note here:
> >
> >* checking a "rcv" member of a DSA-specific structure to anwser the
> >  question "does this interface uses hardware-inserted tag?" is not
> >  generic and not robust at all.
> >
> >* the "dsa_ptr" of net_device is just used to access the dsa_switch_tree
> >  from DSA packet_type receive functions. There must be another way to
> >  pass it, maybe from a netdev_priv or the packet_type->af_packet_priv?
> 
> I sent previously patch for this:
> http://patchwork.ozlabs.org/patch/336940/
> So now my patch would have another user :)

Your patch makes sense. It will reduce the bloating of net_device that
David is talking about. I would also suspect that other <protocol>_ptr
members of the structure are only used in the context of packet_type.

> Vivien, I will refresh the patch and send it to you, the you can
> use the priv by dsa and send my patch along with your patchset. How does
> that sound?

Sure, sounds good.

Thanks,
-v