[dpdk-dev] [PATCH dpdk 5/5] RFC: vfio/ppc64/spapr: Use correct bus addresses for DMA map

[Burakov, Anatoly]

Hi Alexey,

> > DPDK tries to allocate all hugepages that are available to find the
> > smallest amount of physically contiguous memory segments to cover the
> > specified memory size. It then releases all those hugepages that it
> > did not need not sure how this is related to orig=1/0 though.
> 
> 
> No, it never does release a single page :-/

That is weird.

As far as I can remember, when EAL initializes the pages, it checks if there are any active locks on hugepage files for a given prefix (which presumably you didn't set, so it uses a default "rte" prefix), and if there aren't, it removes the hugepage files. That way, if the pages are still in use (e.g. by a secondary process), they aren't removed, but if they aren't used, then they are freed, and reserved back.

That is, technically, DPDK never "frees" any pages (unless you don't supply -m/--socket-mem switch, in which case it does free unused pages, but still leaves used pages after exit), so after a DPDK process exit they're not cleaned up. However, whenever a primary DPDK process runs again, it is usually able to clean them up and thus should be able to initialize again. Perhaps something is preventing file removal from your hugetlbfs? Like, maybe a permissions issue or something?

Thanks,
Anatoly