Re: more tcpdumpinfo for nfs3 problem: aix-server --- linux 2.4.15pre5 client

[kuznet@ms2.inr.ac.ru]

Hello!

> You are saying that the it is impossible for sock_alloc_send_skb() to
> fail when using non-blocking writes?

It is possible and normal provided frame is not fragmented.
And this is bug in nfsd if this happens with its frames.


> writes. (Note: by 'simultaneous' I mean that we don't wait for the
> server to reply before firing off the next request)

I do not understand, you have said you wait for write space yet. :-)


> I haven't done anything about this because IMHO it makes more sense to
> have the QDIO driver drop their special spinlock when calling external
> functions such as dev_kfree_skb_any() 

It is pretty normal, if I understand your words correctly.
kfree_skb() is called under various kinds of locks in lots of places.

> rather than to force the RPC layer to use the spin_lock_irqsave().

I see no relation at all. Do it irqsave and nothing will change,
write_space is called only from softirqs.

It is bug in xprt level to grab spinlock which can cause deadlocks
inside write_space. Probably, I misunderstood you.

Alexey