Re: [hpoj-devel] SMP patches?

[Burkhard K. <bu...@bu...>]

Robert G. Brown
> On Tue, 5 Dec 2000, PASCHAL,DAVID (HP-Roseville,ex1) wrote:
> 
> > Hi, Robert.  It also sounds like an SMP problem to me.  Another thing you
> > could try is to boot the SMP box with the "nosmp" kernel parameter, which
> > puts it into UP mode.  I'll attempt to look into the problem in the near
> > future, but I'll have to dig through a part of the code I'm not currently
> > very familiar with.
> 

Hi Robert,

as one of those two persons being responsible for the 2.4/SMP patch
(although I don't have an SMP box and not very much SMP expertise)
I feel obliged to look into your problem. For now I do not understand
the nature of it - it seems the sendbuffer gets filled without
mlcpp_transmit knowing about it. That's strange because the only
place where sendbuf gets filled is mlcpp_transmit. But we have 
couple of 
	kfree_skb(l->sendbuf); l->sendbuf = NULL;
instructions in different functions. Could there be a need to
make this instruction pair atomic? 

Do you know whether there is a way to arrange for the debug output 
to include CPU-identification?

2.2.16 was known to have some problems. I don't recall wether those
probs where SMP related, but anyhow - would it be possible for
you to try 2.2.17? We have a couple of "works"-reports on 2.2.17
kernels. 


Damcha, any ideas?

Burkhard
-- 
Burkhard Kohl			bu...@bu...