On Wed, Sep 07, 2005 at 11:19:42AM -0700, Stephen Hemminger wrote: > On Wed, 7 Sep 2005 10:19:33 -0700 > Matt Mackall <[EMAIL PROTECTED]> wrote: > > > On Tue, Sep 06, 2005 at 09:02:42PM -0700, Eugene Surovegin wrote: > > > I don't quite understand, you _already_ have deferred processing in > > > netpoll (btw, how good this will work with kgdboe?). > > > > Yes, it exists as a last ditch fallback for cases where we can detect > > deadlock. And it works not at all for kgdboe. Fortunately the fallback > > case rarely coincides with a breakpoint unless you're trying to debug > > the network stack itself (which is asking for trouble). > > > > > If some drivers cannot handle *additional* restriction (nowhere > > > documented, btw), ok, let's have a fallback mode for them, which you > > > _already_ have. > > > > If we have to use the fallback mode all the time on a particular device, > > netpoll has no value on that device. It can't catch the oopses you can't > > already catch most of with syslogd, and it can't be used for kgdb. > > At that point, why bother? > > > > > It's amazing, frankly, you insist that this feature "works just fine", > > > > I have in fact said since the first message in the thread that you > > guys are right, I'm breaking the rules. But I also claim it works fine > > for some hardware. These are not incompatible statements. The second > > is a reminder that this stuff is actually useful to people in its > > current form and that we shouldn't break that. > > > > It really sounds like netconsole needs to have a different device hook > instead of start_xmit. It also probably doesn't want to have allocate > an skb. What you want is a synchronous way to send one packet with > interrupts disabled: > > (dev->netpoll_send)(dev, packet, len) > > Yes, it means fixing lots of device drivers, or maybe a transitional > stub can be set in register_netdevice that uses start_xmit. That is > unsafe, but can used while you go about fixing the other devices.
Netpoll already keeps a pool of skbs, so that's less of a problem. An interface like the above would be fine, and I could use hard_start_xmit for devices that haven't implemented it yet. -- Mathematics is the supreme nostalgia of our time. - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
