Re: [tipc-discussion] send_msg() may not be waken up by tipc_recv_msg() when port congestion happen

[Stephens, A. <all...@wi...>]

Hi Jason:

I don't think the problem you have identified can actually occur because TIPC's link "aliveness" checking ensures that tipc_recv_msg() gets invoked at regular intervals. (That is, a link endpoint must periodically receive either payload messages or LINK_PROTOCOL messages from its peer endpoint, or else the link is declared dead and gets reset.) So even if a sending thread adds its port to a link's list of waiting ports after tipc_recv_msg() has marked the link as non-congested and then sleeps, the next invocation of tipc_recv_msg() will wake the sleeping thread up and allow execution to continue.

Regards,
Al

> -----Original Message-----
> From: jason [mailto:huz...@gm...]
> Sent: Thursday, March 24, 2011 10:54 AM
> To: tip...@li...
> Subject: [tipc-discussion] send_msg() may not be waken up by
> tipc_recv_msg() when port congestion happen
> 
> Hi All,
> 
> When I studying tipc-1.7.6 code. I see a potaintial issue that when a user
> thread which calls send_msg() and is scheduled out because port congestion,
> it may not be wake up any more by tipc_recv_msg().
> 
> Let's see send_msg related pseudo code:
> 01 add self to link.sleep_queue
> 02 if port.congest == 1 {
> 03      schedule();
> 04 }
> 
> and see tipc_recv_msg related code:
> 
> 01 if link.sleep_queue empty {
> 02     port.congest = 0;
> 03     wakeup;
> 04 }
> Say we are SMP. Right after CPU0 runs at tipc_recv_msg related code line 01
> and see link.sleep_queue IS empty, CPU1 execute  send_msg related code
> from line 01 to 04. In this conditon, send_msg will be schedule out by CPU1
> and
> CPU0 loss the opptunity to wake it up. If there is no incoming message any
> more, then send_msg thread will not be waken any more.
> 
> If my analysis is right, I think one way to solve whis issue is to swap
> tipc_recv_msg related code line 01 and 02, and smp_wmb should also be
> placed between them.  But real code change is not as easy as a code line
> swapping.
> 
> Thanks!
> 
> --
> Yours,
> jason
> ------------------------------------------------------------------------------
> Enable your software for Intel(R) Active Management Technology to meet
> the growing manageability and security demands of your customers.
> Businesses are taking advantage of Intel(R) vPro (TM) technology - will your
> software be a part of the solution? Download the Intel(R) Manageability
> Checker today! http://p.sf.net/sfu/intel-dev2devmar
> _______________________________________________
> tipc-discussion mailing list
> tip...@li...
> https://lists.sourceforge.net/lists/listinfo/tipc-discussion