[rds-devel] Fw: Using interrupt context

Dirk Kirsten dirk.kirsten at de.ibm.com
Thu May 20 05:38:11 PDT 2010


I just noticed that I accidentally sent this message to Andy in private. In
fact, it belongs to the rds mailing list.
Another point I want to mention is that this is my bachelor thesis project.
So what happens to my code (the distribution) is not really influenceable
by me and the transport module I am implementing is kind of a research
project, so I do not know if it really will be included (and maintained) by
IBM.

Anyways, here is the message:

Hi,

Actually I already implemented it using bottom-halves. So it is in fact a
good idea to give you some performance numbers. I will try to do this the
next few weeks (could take some time because there is also other stuff to
do...) and I will give you the results as soon as I got them.

Anyway, I do not know if the code is going to be submitted. The lawyers
will have to decide this... (in my personal opinion it should, of course!)

I want to point out again the issue about the gfp value. In my opinion this
is not strictly related to my implementation, I simply do not get why the
gfp value from __rds_conn_create() is not passed to underlying functions.
At the moment the gfp value is useless in my opinion - even if I set it to
GFP_ATOMIC it will still be GFP_KERNEL in rds_conn_from_addr() and so the
gfp in __rds_conn_create() seems to be not usable. If there is a reason why
you are doing it this way, I would be eager to know it.

Mit freundlichen Grüßen / Kind regards,

Dirk Kirsten

----------------------------------------------------------------------
IBM Deutschland GmbH

Geschäftsführung: Martin Jetter (Vorsitzender),
Reinhard Reschke, Christoph Grandpierre, Matthias Hartmann, Michael Diemer,
Martina Koederitz
Sitz der Gesellschaft: Ehningen / Registergericht: Amtsgericht Stuttgart,
HRB 24938


                                                                                                                                    
  From:       Andy Grover <andy.grover at oracle.com>                                                                                  
                                                                                                                                    
  To:         rds-devel at oss.oracle.com                                                                                              
                                                                                                                                    
  Date:       17.05.2010 20:37                                                                                                      
                                                                                                                                    
  Subject:    Re: [rds-devel] Using interrupt context                                                                               
                                                                                                                                    





On 05/17/2010 12:52 AM, Dirk Kirsten wrote:
> I am currently developing a transport module for RDS. There seems to be a
> problem with the synchronization in the current code. I would like to
call
> rds_conn_create and the whole receiving part from within the interrupt
> context. If I do so, the GFP_KERNEL in rds_conn_from_addr() seems to be a
> problem. In __rds_conn_create there is still a gfp_t passed as argument,
so
> would it be possible to simply pass this to rds_conn_from_addr() as well?
> Or is there any reason why you are not doing this?

Hi Dirk,

It's great to hear RDS is proving useful to others.

If you want rds core changes to support your code, I infer you're
planning on submitting it for upstream inclusion? If so, I'd recommend
doing that first, even if the performance is not optimized. We're much
more likely to accept rds core changes for in-tree code, especially
since it sounds like you're doing something a little bit different than
vanilla RDS. We'd like to fully grasp the implications, you know?

Thanks -- Regards -- Andy

_______________________________________________
rds-devel mailing list
rds-devel at oss.oracle.com
http://oss.oracle.com/mailman/listinfo/rds-devel







More information about the rds-devel mailing list