Re: Re: READ_LOCK?

Re: Re: READ_LOCK?

[Harald Welte]

[-- Attachment #1: Type: text/plain, Size: 1155 bytes --]

On Wed, Jan 07, 2004 at 09:27:14AM +0800, zhaoyao wrote:
> Harald Welte,??????
> 
> Yes,as you know in 2.4 or 2.6, bh is similar to softirq,
> WRITE_LOCK is write_lock_bh(&a)
> if u write_lock_bh(&a), obviously other cpu can't enter this critical region,

I don't really get your point.  bh is not really that similar to
softirq.  As stated before, their properties wrt. serialization are
quite different.  the term 'bh' in spin_lock_bh is only a legacy name.
In 2.6.x it mainly deals with disabling preemption.

> but already in softirq, why lock bh again. softirq can't reenter.
> so lock bh is not useful.

As indicated in my last email, I think you are misunderstanding the
kernel contexts.  the network rx softirq _will_ run simultaneously on
multiple cpu's.

-- 
- Harald Welte <laforge@netfilter.org>             http://www.netfilter.org/
============================================================================
  "Fragmentation is like classful addressing -- an interesting early
   architectural error that shows how much experimentation was going
   on while IP was being designed."                    -- Paul Vixie

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Re: READ_LOCK?

[zhaoyao]

Harald Welte,您好！

Yes,as you know in 2.4 or 2.6, bh is similar to softirq,
WRITE_LOCK is write_lock_bh(&a)
if u write_lock_bh(&a), obviously other cpu can't enter this critical region,
but already in softirq, why lock bh again. softirq can't reenter.
so lock bh is not useful.
>On Sun, Jan 04, 2004 at 02:53:03PM +0800, zhaoyao wrote:
>> netfilter-devel,hello!
>> In init_conntrack, why will READ_LOCK or WRITE_LOCK, we'are already in
>> bh, why lock it again?
>
>ip_conntrack is part of netfilter, which in turn is part of the 2.4.x
>and 2.6.x kernel.  Both 2.4 and 2.6 kernel series don't have a 'bottom
>half' for the network stack anymore.  Instead, softirq's are used.
>Softirq's are fully parallelized, i.e. they can run on any number of
>cpu's simultaneously.
>
>> ????????????????????2004-01-04
>
>--
>- Harald Welte <laforge@netfilter.org>             http://www.netfilter.org/
>============================================================================
>  "Fragmentation is like classful addressing -- an interesting early
>   architectural error that shows how much experimentation was going
>   on while IP was being designed."                    -- Paul Vixie

= = = = = = = = = = = = = = = = = = = =
			

　　　　　　　　致
礼！

				
　　　　　　　　zhaoyao
　　　　　　　　zhaoyao@harbournetworks.com
　　　　　　　　　　2004-01-07