On Fri, Oct 08, 2004 at 09:06:10AM +0200, Colin Leroy wrote:
> On 07 Oct 2004 at 23h10, Andi Kleen wrote:
>
> Hi,
>
> > > This patch should do that. It works OK for me, but I'd like it
> > > checked before sent upstream...
> > >
> > > However, it doesn't fix the hang. it looks like this hang is really
> > > coming from sungem.
> >
> > IMHO it's not needed. Taking xmit_lock is harmless even when
> > the NETIF_F_LLTX flag is set.
>
> Should that be completely dropped, or is it still ok ? (I think
> differenciating action based on hard_start_xmit status, that is, don't
> goto repeat undefinitely when NETDEV_TX_BUSY, could be a good idea).
> I mean, should I rework that patch, forget about it or leave it as-is?
Well the purpose of the LLTX flag is to reduce serializing on the
xmit_lock. If we take the lock anyway, that should be harmless as Andi
says. So I'm afraid it looks to be a performance fix at best (which is
a low priority here). Let's back burner it for now.
> Concerning the hang, I see that Andrew has put my first patch, the one
> checking for netif_carrier_ok(), in his tree. Is it an OK solution from
> your (net dev hackers) point of view?
It seems to be papering over a driver bug of some sort, which is not
the way we like to fix things.
--
Mathematics is the supreme nostalgia of our time.
|