[LITMUS^RT] RFP: Jumbo Ring Buffers

[Glenn Elliott]

On Apr 23, 2013, at 12:18 PM, Christopher Kenna <cjk at cs.unc.edu> wrote:

> On Mon, Apr 22, 2013 at 11:55 PM, Björn Brandenburg <bbb at mpi-sws.org> wrote:
>> Thanks a lot for the pointers! How about this function?
>> 
>>        http://lxr.linux.no/linux+v3.0/mm/vmalloc.c#L1680
>> 
>> From the inline documentation, it doesn't sound like there's a limit on the maximum vmalloc() allocation. We don't care whether it is expensive or not; it's a one-time thing anyways.
> 
> I'm no expert, but it seems reasonable. It looks like it does
> something similar (eventually calling map_vm_area, like the other
> function), but also calls kzalloc/kmalloc so that you don't have to
> worry about allocating/tracking your own pages. Glenn might be
> interested in the 'node' parameter, too.

As far as numa is concerned, per-CPU buffers should be local.  Singleton buffers should be interleaved evenly across all nodes.

Speaking of numa, I may have numa-related updates for liblitmus in a few days.

In other news, I used kdump for the first time yesterday.  It will completely change how I develop Litmus outside of a virtual machine.  How has this not been a part of our standard development process?

-Glenn