[Click] Kernel BufferQueue

[Beyers Cronje]

Thanks Eddie,

I only came across CLICK_LALLOC after my original post while looking at
DeQueue. Obviously your post below is more elegant and the right way to go.

Slowly but surely I'm getting to these undocumented features :)

Beyers



On 9/19/06, Eddie Kohler <kohler at cs.ucla.edu> wrote:
>
> Hi Beyers,
>
> So Click actually supports a type of allocation that dynamically decides
> whether to use kmalloc() or vmalloc(), based on the size of the data to be
> allocated.  The allocation and freeing functions are
>
>    void *ptr = CLICK_LALLOC(size_t size);
>    CLICK_LFREE(void *ptr, size_t size);
>
> Note that you have to specify the size when you delete the object as well
> as
> when you allocate it.
>
> I've gone ahead and changed the Queue-derived elements to use LALLOC and
> LFREE
> instead of new[] and delete[].  That should mean that you can use regular
> Queues with huge capacities, no need for a special element.
>
> Thanks,
> Eddie
>
>
> Beyers Cronje wrote:
> > Hi Eddie, all,
> >
> > I had a need to buffer incoming packets in Click kernel module for two
> > seconds on an interface with around 350+mbps traffic load with an
> > average packet count of 80,000pps. In other words I needed to buffer at
> > least 160,000 packets at any given time in the kernel module. Any
> > SimpleQueue derived queue's maximum capacity is 32761, limited by
> > kmalloc's maximum bucket size of 131048 bytes.  Click-install will crash
> > the kernel if you specify any larger capacity.
> >
> > The attached element is a linuxmodule only element that uses vmalloc
> > instead of kmalloc to allocate queue memory. It installs and seems to
> > run fine on my system with a queue capacity of 500,000. It is derived
> > from NotifierQueue and overrides methods used in SimpleQueue to allocate
> > and free memory (initialize, live_reconfigure & cleanup).
> >
> > Feel free to use or include in Click. Also feel free to rename as I
> > couldnt come up with a better name than BufferQueue :)
> >
> > One drawback of calling vmalloc directly is that click/meminfo will not
> > reflect this allocation. Maybe a Click newvm(size_t) function can help
> > with this.
> >
> > Cheers
> >
> > Beyers Cronje
>