On 14:03 Sun 23 Nov     , Yevgeny Kliteynik wrote:
> Sasha,
>
> Sasha Khapyorsky wrote:
>> Hi Yevgeny,
>> On 14:46 Thu 20 Nov     , Yevgeny Kliteynik wrote:
>>> I can do something like the following patch, but I have
>>> some strange feeling that I'm missing something...
>> I cannot see any errors here. But probably you can use simpler approach
>> - just cleanup all switch's lft_buf separately after ucast_mgr is
>> finished (including wait_for_pending_transactions()).
>
> I've been doing some thinking...
> Basically, what you're saying is that at the end of each and
> every heavy sweep you will free ALL the lft_buf arrays, unless
> there was some error, that should trigger a new heavy sweep
> anyway. So what's the point of having lft_buf in the first place?
>
> It was relevant in the beginning of ucast cache implementation,
> but now after all the lft simplifications, I don't see how it
> is used. Am I missing something here, or should we just remove
> all these lft_buf and go back to single ucast_mgr_t.lft_buf?

As far as I remember it was your idea to use newly generated lft_buf in
cache regarless to the state of current LFTs. No?

Also we have strange bug#1406 yet...

Sasha
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to