From: Jarek Poplawski

Date: Mon, 18 Aug 2008 00:22:07 +0200
> Actually, I've only asked you to withdraw this patch for now, but I'm

quoted text
> still not convinced you're right. You should better show me first the

> place where this can make a difference. (I think this test broke for

> some other non RCU reason.) So, maybe you're right, but I've to check

> this more.

It _might_ be ok once we are done sorting out the synchronization

sequence dev_deactivate() uses.  But we aren't there yet.
> BTW, I guess you've seen this other thread: "panic 2.6.27-rc3-git2,

quoted text
> qdisc_dequeue_head" where Denys and I fight with this new locking.

> Alas, it looks to me as a real mess, and I currently try with this

> previous idea of netdev_queue->qdisc_lock, which you didn't like too.

> But, after looking at the current bugs shown by debugging I really

> think we'll have bugs here all the time without simplifying this.

> I think my concept should work soon, but if you don't agree with

> this at all we can stop and wait for better ideas.

I can't even follow your flurry of patches, and neither can the

tester :-)  I deleted the entire thread to be honest, hoping you

would come back with a simple analysis once you've worked things

out with the tester.
What is the real problem besides the correct notify_and_destroy()

issue you discovered?
The locking we have now is very simple:
1) Only under RTNL can qdisc roots change.
2) Therefore, sch_tree_lock() and tcf_tree_lock() are fully valid

   and lock the entire qdisc tree state, if and only if used under

   RTNL lock.
3) Before modifying a qdisc, we dev_deactivate(), which synchronizes

   with asynchronous TX/RX packet handling contexts.
4) The qdisc root and all children are protected by the root qdiscs

   lock, which is taken when asynchonous contexts need to blocked

   while modifying some root or inner qdisc's state.
Yes, of course, if you apply a hammer and add a bit lock at the

top of all of this it will fix whatever bugs remain, but as you

know I don't think that's the solution.
The only substance I've seen is that you've found a violation of #4 in

notify_and_destroy(), so great let's test the fix for that.

--

To unsubscribe from this list: send the line "unsubscribe netdev" in

the body of a message to majordomo@vger.kernel.org

More majordomo info at  http://vger.kernel.org/majordomo-info.html